From e238dd724ac62554d5bc487fc29820227705de80 Mon Sep 17 00:00:00 2001
From: David Jurado <david.jurado@factored.ai>
Date: Fri, 23 Jul 2021 07:57:30 -0500
Subject: [PATCH 01/14] Add download data task to MLCube

---
 object_detection/.dockerignore                |   2 +
 object_detection/.gitignore                   |   1 +
 object_detection/Dockerfile.mlcube            |  31 ++++
 object_detection/README.md                    | 136 ++++++------------
 object_detection/README_OLD.md                |  95 ++++++++++++
 object_detection/download_dataset.sh          |  25 ++--
 object_detection/mlcube.py                    |  62 ++++++++
 object_detection/mlcube/mlcube.yaml           |  32 +++++
 object_detection/mlcube/mlcube_cli.py         |  70 +++++++++
 .../mlcube/workspace/parameters.yaml          |   0
 10 files changed, 351 insertions(+), 103 deletions(-)
 create mode 100644 object_detection/.dockerignore
 create mode 100644 object_detection/.gitignore
 create mode 100644 object_detection/Dockerfile.mlcube
 create mode 100644 object_detection/README_OLD.md
 create mode 100644 object_detection/mlcube.py
 create mode 100644 object_detection/mlcube/mlcube.yaml
 create mode 100644 object_detection/mlcube/mlcube_cli.py
 create mode 100644 object_detection/mlcube/workspace/parameters.yaml

diff --git a/object_detection/.dockerignore b/object_detection/.dockerignore
new file mode 100644
index 000000000..3d19cfc1d
--- /dev/null
+++ b/object_detection/.dockerignore
@@ -0,0 +1,2 @@
+# Do not add MLCube's workspace directory.
+mlcube/
\ No newline at end of file
diff --git a/object_detection/.gitignore b/object_detection/.gitignore
new file mode 100644
index 000000000..eda0ebdf5
--- /dev/null
+++ b/object_detection/.gitignore
@@ -0,0 +1 @@
+mlcube/workspace/data
\ No newline at end of file
diff --git a/object_detection/Dockerfile.mlcube b/object_detection/Dockerfile.mlcube
new file mode 100644
index 000000000..b277b60ab
--- /dev/null
+++ b/object_detection/Dockerfile.mlcube
@@ -0,0 +1,31 @@
+FROM pytorch/pytorch:1.6.0-cuda10.1-cudnn7-devel
+
+RUN echo 'debconf debconf/frontend select Noninteractive' | debconf-set-selections
+
+# install basics
+RUN apt-get update -y \
+ && apt-get install -y apt-utils curl unzip pv \
+                       libglib2.0-0=2.56.1-2ubuntu1 \
+                       libsm6=2:1.2.2-1 \
+                       libxext6=2:1.3.3-1 \
+                       libxrender-dev=1:0.9.10-1
+
+RUN pip install ninja==1.8.2.post2 \
+                yacs==0.1.5 \
+                cython==0.29.5 \
+                matplotlib==3.0.2 \
+                opencv-python==4.0.0.21 \
+                mlperf_compliance==0.0.10 \
+                torchvision==0.2.2 \
+                pycocotools==2.0.2 \
+                typer==0.3.2
+
+RUN pip install --no-cache-dir https://github.com/mlperf/logging/archive/9ea0afa.zip
+
+WORKDIR /workspace
+
+COPY . .
+
+RUN chmod +x ./run_and_time.sh ./download_dataset.sh
+
+ENTRYPOINT ["python", "/workspace/mlcube.py"]
\ No newline at end of file
diff --git a/object_detection/README.md b/object_detection/README.md
index 6895735de..8d40ef0c6 100644
--- a/object_detection/README.md
+++ b/object_detection/README.md
@@ -1,95 +1,41 @@
-# 1. Problem
-Object detection and segmentation. Metrics are mask and box mAP.
-
-# 2. Directions
-
-### Steps to configure machine
-
-1. Checkout the MLPerf repository
-```
-mkdir -p mlperf
-cd mlperf
-git clone https://github.com/mlperf/training.git
-```
-2. Install CUDA and Docker
-```
-source training/install_cuda_docker.sh
-```
-3. Build the docker image for the object detection task
-```
-cd training/object_detection/
-nvidia-docker build . -t mlperf/object_detection
-```
-
-4. Run docker container and install code
-```
-nvidia-docker run -v .:/workspace -t -i --rm --ipc=host mlperf/object_detection \
-    "cd mlperf/training/object_detection && ./install.sh"
-```
-Now exit the docker container (Ctrl-D) to get back to your host.
-
-### Steps to download data
-```
-# From training/object_detection/
-source download_dataset.sh
-```
-
-### Steps to run benchmark.
-```
-nvidia-docker run -v .:/workspace -t -i --rm --ipc=host mlperf/object_detection \
-    "cd mlperf/training/object_detection && ./run_and_time.sh"
-```
-
-# 3. Dataset/Environment
-### Publication/Attribution
-Microsoft COCO: Common Objects in Context
-
-### Data preprocessing
-Only horizontal flips are allowed.
-
-### Training and test data separation
-As provided by MS-COCO (2017 version).
-
-### Training data order
-Randomly.
-
-### Test data order
-Any order.
-
-# 4. Model
-### Publication/Attribution
-He, Kaiming, et al. "Mask r-cnn." Computer Vision (ICCV), 2017 IEEE International Conference on.
-IEEE, 2017.
-
-We use a version of Mask R-CNN with a ResNet50 backbone.
-
-### List of layers
-Running the timing script will display a list of layers.
-
-### Weight and bias initialization
-The ResNet50 base must be loaded from the provided weights. They may be quantized.
-
-### Loss function
-Multi-task loss (classification, box, mask). Described in the Mask R-CNN paper.
-
-Classification: Smooth L1 loss
-
-Box: Log loss for true class.
-
-Mask: per-pixel sigmoid, average binary cross-entropy loss.
-
-### Optimizer
-Momentum SGD. Weight decay of 0.0001, momentum of 0.9.
-
-# 5. Quality
-### Quality metric
-As Mask R-CNN can provide both boxes and masks, we evaluate on both box and mask mAP.
-
-### Quality target
-Box mAP of 0.377, mask mAP of 0.339
-
-### Evaluation frequency
-Once per epoch, 118k.
-
-### Evaluation thoroughness
-Evaluate over the entire validation set. Use the official COCO API to compute mAP.
+## Current implementation
+
+We'll be updating this section as we merge MLCube PRs and make new MLCube releases.
+
+### Project setup
+```Python
+# Create Python environment 
+virtualenv -p python3 ./env && source ./env/bin/activate
+
+# Install MLCube and MLCube docker runner from GitHub repository (normally, users will just run `pip install mlcube mlcube_docker`)
+git clone https://github.com/sergey-serebryakov/mlbox.git && cd mlbox && git checkout feature/configV2
+cd ./runners/mlcube_docker && export PYTHONPATH=$(pwd)
+cd ../../ && pip install -r mlcube/requirements.txt && pip install omegaconf && cd ../
+
+# Fetch the RNN speech recognition workload
+git clone https://github.com/mlcommons/training && cd ./training
+git fetch origin pull/491/head:feature/object_detection && git checkout feature/object_detection
+cd ./object_detection/mlcube
+```
+
+### Dataset
+
+
+The COCO dataset will be downloaded and extracted. Sizes of the dataset in each step:
+
+| Dataset Step                   | MLCube Task       | Format         | Size     |
+|--------------------------------|-------------------|----------------|----------|
+| Download (Compressed dataset)  | download_data     | Tar/Zip files  | ~20.5 GB |
+| Extract (Uncompressed dataset) | download_data     | Jpg/Json files | ~21.2 GB |
+| Total                          | (After all tasks) | All            | ~41.7 GB |
+
+### Tasks execution
+```
+# Download COCO dataset. Default path = /workspace/data
+# To override it, use --data_dir=DATA_DIR
+python mlcube_cli.py run --task download_data --platform docker
+
+# Run benchmark. Default paths = ./workspace/data
+# Parameters to override: --data_dir=DATA_DIR, --output_dir=OUTPUT_DIR, --parameters_file=PATH_TO_TRAINING_PARAMS
+python mlcube_cli.py run --task train --platform docker
+```
\ No newline at end of file
diff --git a/object_detection/README_OLD.md b/object_detection/README_OLD.md
new file mode 100644
index 000000000..6895735de
--- /dev/null
+++ b/object_detection/README_OLD.md
@@ -0,0 +1,95 @@
+# 1. Problem
+Object detection and segmentation. Metrics are mask and box mAP.
+
+# 2. Directions
+
+### Steps to configure machine
+
+1. Checkout the MLPerf repository
+```
+mkdir -p mlperf
+cd mlperf
+git clone https://github.com/mlperf/training.git
+```
+2. Install CUDA and Docker
+```
+source training/install_cuda_docker.sh
+```
+3. Build the docker image for the object detection task
+```
+cd training/object_detection/
+nvidia-docker build . -t mlperf/object_detection
+```
+
+4. Run docker container and install code
+```
+nvidia-docker run -v .:/workspace -t -i --rm --ipc=host mlperf/object_detection \
+    "cd mlperf/training/object_detection && ./install.sh"
+```
+Now exit the docker container (Ctrl-D) to get back to your host.
+
+### Steps to download data
+```
+# From training/object_detection/
+source download_dataset.sh
+```
+
+### Steps to run benchmark.
+```
+nvidia-docker run -v .:/workspace -t -i --rm --ipc=host mlperf/object_detection \
+    "cd mlperf/training/object_detection && ./run_and_time.sh"
+```
+
+# 3. Dataset/Environment
+### Publication/Attribution
+Microsoft COCO: Common Objects in Context
+
+### Data preprocessing
+Only horizontal flips are allowed.
+
+### Training and test data separation
+As provided by MS-COCO (2017 version).
+
+### Training data order
+Randomly.
+
+### Test data order
+Any order.
+
+# 4. Model
+### Publication/Attribution
+He, Kaiming, et al. "Mask r-cnn." Computer Vision (ICCV), 2017 IEEE International Conference on.
+IEEE, 2017.
+
+We use a version of Mask R-CNN with a ResNet50 backbone.
+
+### List of layers
+Running the timing script will display a list of layers.
+
+### Weight and bias initialization
+The ResNet50 base must be loaded from the provided weights. They may be quantized.
+
+### Loss function
+Multi-task loss (classification, box, mask). Described in the Mask R-CNN paper.
+
+Classification: Smooth L1 loss
+
+Box: Log loss for true class.
+
+Mask: per-pixel sigmoid, average binary cross-entropy loss.
+
+### Optimizer
+Momentum SGD. Weight decay of 0.0001, momentum of 0.9.
+
+# 5. Quality
+### Quality metric
+As Mask R-CNN can provide both boxes and masks, we evaluate on both box and mask mAP.
+
+### Quality target
+Box mAP of 0.377, mask mAP of 0.339
+
+### Evaluation frequency
+Once per epoch, 118k.
+
+### Evaluation thoroughness
+Evaluate over the entire validation set. Use the official COCO API to compute mAP.
diff --git a/object_detection/download_dataset.sh b/object_detection/download_dataset.sh
index e9fe2656d..b05d180fc 100755
--- a/object_detection/download_dataset.sh
+++ b/object_detection/download_dataset.sh
@@ -1,20 +1,29 @@
 #!/bin/bash
 
 # Get COCO 2017 data sets
-mkdir -p pytorch/datasets/coco
-pushd pytorch/datasets/coco
+DATA_ROOT_DIR="${DATA_ROOT_DIR:-./pytorch/datasets/coco}"
+echo "Downloaading to folder: $DATA_ROOT_DIR"
+mkdir -p $DATA_ROOT_DIR
+pushd $DATA_ROOT_DIR
 
 curl -O https://dl.fbaipublicfiles.com/detectron/coco/coco_annotations_minival.tgz
-tar xzf coco_annotations_minival.tgz
+echo "Extracting coco_annotations_minival.tgz ..."
+tar -xzf coco_annotations_minival.tgz &>/dev/null
 
-curl -O http://images.cocodataset.org/zips/train2017.zip
-unzip train2017.zip
+curl -O http://images.cocodataset.org/annotations/annotations_trainval2017.zip
+echo "Extracting annotations_trainval2017.zip ..."
+n_files=`unzip -l  annotations_trainval2017.zip| grep .json | wc -l`
+unzip annotations_trainval2017.zip | pv -l -s $n_files > /dev/null
 
 curl -O http://images.cocodataset.org/zips/val2017.zip
-unzip val2017.zip
+echo "Extracting val2017.zip ..."
+n_files=`unzip -l  val2017.zip| grep .jpg | wc -l`
+unzip val2017.zip | pv -l -s $n_files > /dev/null
 
-curl -O http://images.cocodataset.org/annotations/annotations_trainval2017.zip
-unzip annotations_trainval2017.zip
+curl -O http://images.cocodataset.org/zips/train2017.zip
+echo "Extracting train2017.zip ..."
+n_files=`unzip -l  train2017.zip| grep .jpg | wc -l`
+unzip train2017.zip | pv -l -s $n_files > /dev/null
 
 # TBD: MD5 verification
 # $md5sum *.zip *.tgz
diff --git a/object_detection/mlcube.py b/object_detection/mlcube.py
new file mode 100644
index 000000000..379818b72
--- /dev/null
+++ b/object_detection/mlcube.py
@@ -0,0 +1,62 @@
+"""MLCube handler file"""
+import os
+import yaml
+import typer
+import shutil
+import subprocess
+from pathlib import Path
+
+
+app = typer.Typer()
+
+class DownloadDataTask(object):
+    """Download task Class
+    It defines the environment variables:
+        DATA_ROOT_DIR: Directory path to download the dataset
+    Then executes the download script"""
+    @staticmethod
+    def run(data_dir: str) -> None:
+
+        env = os.environ.copy()
+        env.update({
+            'DATA_ROOT_DIR': data_dir,
+        })
+
+        process = subprocess.Popen("./download_dataset.sh", cwd=".", env=env)
+        process.wait()
+
+class TrainTask(object):
+    """Preprocess dataset task Class
+    It defines the environment variables:
+        DATA_DIR: Dataset directory path
+        OUTPUT_DIR: Directory path where model will be saved
+        All other parameters defined in parameters_file
+    Then executes the benchmark script"""
+    @staticmethod
+    def run(data_dir: str, output_dir: str, parameters_file: str) -> None:
+        with open(parameters_file, 'r') as stream:
+            parameters = yaml.safe_load(stream)
+
+        env = os.environ.copy()
+        env.update({
+            'DATA_DIR': data_dir,
+            'OUTPUT_DIR': output_dir,
+        })
+
+        env.update(parameters)
+
+        process = subprocess.Popen("./run_and_time.sh", cwd=".", env=env)
+        process.wait()
+
+@app.command("download_data")
+def download_data(data_dir: str = typer.Option(..., '--data_dir')):
+    DownloadDataTask.run(data_dir)
+
+@app.command("train")
+def train(data_dir: str = typer.Option(..., '--data_dir'),
+          output_dir: str = typer.Option(..., '--output_dir'),
+          parameters_file: str = typer.Option(..., '--parameters_file')):
+    TrainTask.run(data_dir, output_dir, parameters_file)
+
+if __name__ == '__main__':
+    app()
\ No newline at end of file
diff --git a/object_detection/mlcube/mlcube.yaml b/object_detection/mlcube/mlcube.yaml
new file mode 100644
index 000000000..eed17a04f
--- /dev/null
+++ b/object_detection/mlcube/mlcube.yaml
@@ -0,0 +1,32 @@
+name: Object Detection
+description: MLCommons Object Detection Training Reference Benchmark
+authors: 
+ - {name: "MLCommons Best Practices Working Group"}
+
+platform:
+  # Edit this according to your system specs
+  accelerator_count: 1
+
+container:
+  # Image name.
+  image: mlcommons/object_detection:0.0.1
+  # Docker build context relative to $MLCUBE_ROOT. Default is `build`.
+  build_context: "../"
+  # Docker file name within docker build context, default is `Dockerfile`.
+  build_file: "Dockerfile.mlcube"
+
+tasks:
+  # Download LibriSpeech dataset
+  download_data:
+    io:
+      # Directory for uncompressed datasets. Total size is ~ 65G.
+      - {name: data_dir, type: directory, io: output, default: $WORKSPACE/data}
+  train:
+  # Train Object Detection model
+    io:
+      # see Download::data_dir, it takes the processes subfolder
+      - {name: data_dir, type: directory, io: input, default: $WORKSPACE/data/LibriSpeech}
+      # Model output folder
+      - {name: output_dir, type: directory, io: output, default: $WORKSPACE/output}
+      # Yaml file with training parameters.
+      - {name: parameters_file, type: file, io: input, default: $WORKSPACE/parameters.yaml}
\ No newline at end of file
diff --git a/object_detection/mlcube/mlcube_cli.py b/object_detection/mlcube/mlcube_cli.py
new file mode 100644
index 000000000..5df6a4bff
--- /dev/null
+++ b/object_detection/mlcube/mlcube_cli.py
@@ -0,0 +1,70 @@
+"""
+This requires the MLCube 2.0 configuration
+"""
+import os
+import yaml
+import click
+import typing
+from mlcube_docker.docker_run import DockerRun
+
+
+def load_config(mlcube_config_path: str, user_config_path: str) -> typing.Dict:
+    """Returns dictionary containing MLCube configuration"""
+    # Load mlcube config data
+    try:
+        with open(mlcube_config_path) as stream:
+            mlcube_config_data = yaml.load(stream.read(), Loader=yaml.SafeLoader)
+    except IOError as exc:
+        # If file doesn't exist throw the exception:
+        # OSError: {PATH_TO}/mnist/mlcube.yaml: No such file or directory
+        raise IOError("%s: %s" % (mlcube_config_path, exc.strerror))
+
+    # Load user config data if file exists
+    if os.path.isfile(user_config_path):
+        with open(user_config_path) as stream:
+            user_config_data = yaml.load(stream.read(), Loader=yaml.SafeLoader)
+    else:
+        return mlcube_config_data
+
+    # Merge config data
+    tmp = mlcube_config_data['container']
+    mlcube_config_data['container'] = user_config_data['container']
+    mlcube_config_data['container'].update(tmp)
+    return mlcube_config_data
+
+
+@click.group(name='mlcube')
+def cli():
+    pass
+
+
+@cli.command(name='run', help='Run MLCube ML task.',
+             context_settings=dict(ignore_unknown_options=True, allow_extra_args=True))
+@click.option('--mlcube', required=False, type=str, help='Path to MLCube directory, default is current.')
+@click.option('--platform', required=False, type=str, help='Platform to run MLCube, default is docker/podman.')
+@click.option('--task', required=False, type=str, help='MLCube task name to run, default is `main`.')
+@click.option('--workspace', required=False, type=str, help='Workspace path, default is `workspace` within '
+                                                            'MLCube folder')
+def run(mlcube: str, platform: str, task: str, workspace: str):
+    mlcube_root = os.path.abspath(mlcube or os.getcwd())
+    if os.path.isfile(mlcube_root):
+        mlcube_root = os.path.dirname(mlcube_root)
+
+    platform = platform or 'docker'
+    if platform != 'docker':
+        raise ValueError(f"Only `docker` platform is supported")
+
+    task = task or 'main'
+    workspace = workspace or os.path.join(mlcube_root, 'workspace')
+
+    mlcube_config_data = load_config(
+        os.path.join(str(mlcube_root), 'mlcube.yaml'),
+        os.path.join(os.path.expanduser("~"), '.mlcube.yaml')
+    )
+
+    docker_runner = DockerRun(mlcube_config_data, root=mlcube_root, workspace=workspace, task=task)
+    docker_runner.run()
+
+
+if __name__ == "__main__":
+    cli()
\ No newline at end of file
diff --git a/object_detection/mlcube/workspace/parameters.yaml b/object_detection/mlcube/workspace/parameters.yaml
new file mode 100644
index 000000000..e69de29bb

From 7f69f94c8263c77e07df2f0a3a7288a76021375d Mon Sep 17 00:00:00 2001
From: David Jurado <david.jurado@factored.ai>
Date: Mon, 26 Jul 2021 11:58:51 -0500
Subject: [PATCH 02/14] Add unzipping progress

---
 object_detection/README.md           |  4 ++--
 object_detection/download_dataset.sh | 18 +++++++++++-------
 2 files changed, 13 insertions(+), 9 deletions(-)

diff --git a/object_detection/README.md b/object_detection/README.md
index 8d40ef0c6..8bd9eb641 100644
--- a/object_detection/README.md
+++ b/object_detection/README.md
@@ -12,9 +12,9 @@ git clone https://github.com/sergey-serebryakov/mlbox.git && cd mlbox && git che
 cd ./runners/mlcube_docker && export PYTHONPATH=$(pwd)
 cd ../../ && pip install -r mlcube/requirements.txt && pip install omegaconf && cd ../
 
-# Fetch the RNN speech recognition workload
+# Fetch the Object Detection workload
 git clone https://github.com/mlcommons/training && cd ./training
-git fetch origin pull/491/head:feature/object_detection && git checkout feature/object_detection
+git fetch origin pull/501/head:feature/object_detection && git checkout feature/object_detection
 cd ./object_detection/mlcube
 ```
 
diff --git a/object_detection/download_dataset.sh b/object_detection/download_dataset.sh
index b05d180fc..fcd857443 100755
--- a/object_detection/download_dataset.sh
+++ b/object_detection/download_dataset.sh
@@ -6,24 +6,28 @@ echo "Downloaading to folder: $DATA_ROOT_DIR"
 mkdir -p $DATA_ROOT_DIR
 pushd $DATA_ROOT_DIR
 
+echo "Downloading coco_annotations_minival.tgz:"
 curl -O https://dl.fbaipublicfiles.com/detectron/coco/coco_annotations_minival.tgz
-echo "Extracting coco_annotations_minival.tgz ..."
+echo "Extracting coco_annotations_minival.tgz:"
 tar -xzf coco_annotations_minival.tgz &>/dev/null
 
+echo "Downloading annotations_trainval2017.zip:"
 curl -O http://images.cocodataset.org/annotations/annotations_trainval2017.zip
-echo "Extracting annotations_trainval2017.zip ..."
+echo "Extracting annotations_trainval2017.zip:"
 n_files=`unzip -l  annotations_trainval2017.zip| grep .json | wc -l`
-unzip annotations_trainval2017.zip | pv -l -s $n_files > /dev/null
+unzip annotations_trainval2017.zip  | { I=-1; while read; do printf "Progress: $((++I*100/$n_files))%%\r"; done; echo ""; }
 
+echo "Downloading val2017.zip:"
 curl -O http://images.cocodataset.org/zips/val2017.zip
-echo "Extracting val2017.zip ..."
+echo "Extracting val2017.zip:"
 n_files=`unzip -l  val2017.zip| grep .jpg | wc -l`
-unzip val2017.zip | pv -l -s $n_files > /dev/null
+unzip val2017.zip  | { I=-1; while read; do printf "Progress: $((++I*100/$n_files))%%\r"; done; echo ""; }
 
+echo "Downloading train2017.zip:"
 curl -O http://images.cocodataset.org/zips/train2017.zip
-echo "Extracting train2017.zip ..."
+echo "Extracting train2017.zip:"
 n_files=`unzip -l  train2017.zip| grep .jpg | wc -l`
-unzip train2017.zip | pv -l -s $n_files > /dev/null
+unzip train2017.zip  | { I=-1; while read; do printf "Progress: $((++I*100/$n_files))%%\r"; done; echo ""; }
 
 # TBD: MD5 verification
 # $md5sum *.zip *.tgz

From a3bc71c010bb76cea7e169fa546fd6d2af50987b Mon Sep 17 00:00:00 2001
From: David Jurado <david.jurado@factored.ai>
Date: Wed, 28 Jul 2021 19:46:41 -0500
Subject: [PATCH 03/14] Add train task

---
 object_detection/.gitignore                   |  3 ++-
 object_detection/Dockerfile.mlcube            |  6 +++--
 object_detection/README.md                    |  2 --
 object_detection/mlcube.py                    |  7 +++++-
 object_detection/mlcube/mlcube.yaml           |  2 +-
 .../mlcube/workspace/parameters.yaml          |  2 ++
 .../maskrcnn_benchmark/engine/trainer.py      |  1 +
 .../pytorch/tools/train_mlperf.py             |  4 ++--
 object_detection/run_and_time.sh              | 24 ++++++++++++++++---
 9 files changed, 39 insertions(+), 12 deletions(-)

diff --git a/object_detection/.gitignore b/object_detection/.gitignore
index eda0ebdf5..7406dc7e9 100644
--- a/object_detection/.gitignore
+++ b/object_detection/.gitignore
@@ -1 +1,2 @@
-mlcube/workspace/data
\ No newline at end of file
+mlcube/workspace/data
+mlcube/workspace/output
\ No newline at end of file
diff --git a/object_detection/Dockerfile.mlcube b/object_detection/Dockerfile.mlcube
index b277b60ab..5fe9dca29 100644
--- a/object_detection/Dockerfile.mlcube
+++ b/object_detection/Dockerfile.mlcube
@@ -4,7 +4,7 @@ RUN echo 'debconf debconf/frontend select Noninteractive' | debconf-set-selectio
 
 # install basics
 RUN apt-get update -y \
- && apt-get install -y apt-utils curl unzip pv \
+ && apt-get install -y apt-utils curl unzip \
                        libglib2.0-0=2.56.1-2ubuntu1 \
                        libsm6=2:1.2.2-1 \
                        libxext6=2:1.3.3-1 \
@@ -26,6 +26,8 @@ WORKDIR /workspace
 
 COPY . .
 
-RUN chmod +x ./run_and_time.sh ./download_dataset.sh
+RUN chmod +x ./run_and_time.sh ./download_dataset.sh ./install.sh 
+
+ENV PYTHONPATH /workspace/pytorch
 
 ENTRYPOINT ["python", "/workspace/mlcube.py"]
\ No newline at end of file
diff --git a/object_detection/README.md b/object_detection/README.md
index 8bd9eb641..e303bce26 100644
--- a/object_detection/README.md
+++ b/object_detection/README.md
@@ -32,10 +32,8 @@ The COCO dataset will be downloaded and extracted. Sizes of the dataset in each
 ### Tasks execution
 ```
 # Download COCO dataset. Default path = /workspace/data
-# To override it, use --data_dir=DATA_DIR
 python mlcube_cli.py run --task download_data --platform docker
 
 # Run benchmark. Default paths = ./workspace/data
-# Parameters to override: --data_dir=DATA_DIR, --output_dir=OUTPUT_DIR, --parameters_file=PATH_TO_TRAINING_PARAMS
 python mlcube_cli.py run --task train --platform docker
 ```
\ No newline at end of file
diff --git a/object_detection/mlcube.py b/object_detection/mlcube.py
index 379818b72..b3f8c2d03 100644
--- a/object_detection/mlcube.py
+++ b/object_detection/mlcube.py
@@ -34,6 +34,10 @@ class TrainTask(object):
     Then executes the benchmark script"""
     @staticmethod
     def run(data_dir: str, output_dir: str, parameters_file: str) -> None:
+
+        process = subprocess.Popen("./install.sh", cwd=".")
+        process.wait()
+
         with open(parameters_file, 'r') as stream:
             parameters = yaml.safe_load(stream)
 
@@ -43,7 +47,8 @@ def run(data_dir: str, output_dir: str, parameters_file: str) -> None:
             'OUTPUT_DIR': output_dir,
         })
 
-        env.update(parameters)
+        if parameters is not None:
+            env.update(parameters)
 
         process = subprocess.Popen("./run_and_time.sh", cwd=".", env=env)
         process.wait()
diff --git a/object_detection/mlcube/mlcube.yaml b/object_detection/mlcube/mlcube.yaml
index eed17a04f..a921adfec 100644
--- a/object_detection/mlcube/mlcube.yaml
+++ b/object_detection/mlcube/mlcube.yaml
@@ -25,7 +25,7 @@ tasks:
   # Train Object Detection model
     io:
       # see Download::data_dir, it takes the processes subfolder
-      - {name: data_dir, type: directory, io: input, default: $WORKSPACE/data/LibriSpeech}
+      - {name: data_dir, type: directory, io: input, default: $WORKSPACE/data}
       # Model output folder
       - {name: output_dir, type: directory, io: output, default: $WORKSPACE/output}
       # Yaml file with training parameters.
diff --git a/object_detection/mlcube/workspace/parameters.yaml b/object_detection/mlcube/workspace/parameters.yaml
index e69de29bb..473864770 100644
--- a/object_detection/mlcube/workspace/parameters.yaml
+++ b/object_detection/mlcube/workspace/parameters.yaml
@@ -0,0 +1,2 @@
+SAVE_CHECKPOINTS: "True" # Instead of False use empty value
+SOLVER_MAX_ITER: "100"
\ No newline at end of file
diff --git a/object_detection/pytorch/maskrcnn_benchmark/engine/trainer.py b/object_detection/pytorch/maskrcnn_benchmark/engine/trainer.py
index 860107246..27a20d7d3 100644
--- a/object_detection/pytorch/maskrcnn_benchmark/engine/trainer.py
+++ b/object_detection/pytorch/maskrcnn_benchmark/engine/trainer.py
@@ -73,6 +73,7 @@ def do_train(
     model.train()
     start_training_time = time.time()
     end = time.time()
+    early_exit = False
 
     for iteration, (images, targets, _) in enumerate(data_loader, start_iter):
 
diff --git a/object_detection/pytorch/tools/train_mlperf.py b/object_detection/pytorch/tools/train_mlperf.py
index 061431459..b7f2b100e 100644
--- a/object_detection/pytorch/tools/train_mlperf.py
+++ b/object_detection/pytorch/tools/train_mlperf.py
@@ -115,7 +115,7 @@ def cast_frozen_bn_to_half(module):
         cast_frozen_bn_to_half(child)
     return module
 
-def train(cfg, local_rank, distributed, disable_allreduce_for_logging, random_number_generator):
+def train(cfg, local_rank, distributed, random_number_generator):
     # Model logging
     log_event(key=constants.GLOBAL_BATCH_SIZE, value=cfg.SOLVER.IMS_PER_BATCH)
     log_event(key=constants.NUM_IMAGE_CANDIDATES, value=cfg.MODEL.RPN.FPN_POST_NMS_TOP_N_TRAIN)
@@ -295,7 +295,7 @@ def main():
         logger.info(config_str)
     logger.info("Running with config:\n{}".format(cfg))
 
-    model, success = train(cfg, args.local_rank, args.distributed, args.disable_allreduce_for_logging, random_number_generator)
+    model, success = train(cfg, args.local_rank, args.distributed, random_number_generator)
 
     if success is not None:
         if success:
diff --git a/object_detection/run_and_time.sh b/object_detection/run_and_time.sh
index f8526534b..f7709d19e 100755
--- a/object_detection/run_and_time.sh
+++ b/object_detection/run_and_time.sh
@@ -1,11 +1,29 @@
 #!/bin/bash
 
-# Runs benchmark and reports time to convergence
 
-pushd pytorch
+DATA_DIR="${DATA_DIR:-pytorch/datasets/coco}"
+DATA_ROOT_TARGET="pytorch/datasets/coco"
+OUTPUT_DIR="${OUTPUT_DIR:-}"
+
+SAVE_CHECKPOINTS="${SAVE_CHECKPOINTS:-}"
+SOLVER_MAX_ITER="${SOLVER_MAX_ITER:-40000}"
 
+#Link input data paths
+if [ "$DATA_DIR" != "$DATA_ROOT_TARGET" ]; then
+       mkdir -p $DATA_ROOT_TARGET
+       ln -s $DATA_DIR/annotations $DATA_ROOT_TARGET/annotations
+       ln -s $DATA_DIR/train2017 $DATA_ROOT_TARGET/train2017
+       ln -s $DATA_DIR/test2017 $DATA_ROOT_TARGET/test2017
+       ln -s $DATA_DIR/val2017 $DATA_ROOT_TARGET/val2017
+       echo $DATA_ROOT_TARGET
+       ls -lah $DATA_ROOT_TARGET
+fi
+
+# Runs benchmark and reports time to convergence
+pushd pytorch
 # Single GPU training
 time python tools/train_mlperf.py --config-file "configs/e2e_mask_rcnn_R_50_FPN_1x.yaml" \
-       SOLVER.IMS_PER_BATCH 2 TEST.IMS_PER_BATCH 1 SOLVER.MAX_ITER 720000 SOLVER.STEPS "(480000, 640000)" SOLVER.BASE_LR 0.0025
+       SOLVER.IMS_PER_BATCH 2 TEST.IMS_PER_BATCH 1 SOLVER.MAX_ITER 720000 SOLVER.STEPS "(480000, 640000)" \
+       SOLVER.MAX_ITER $SOLVER_MAX_ITER SOLVER.BASE_LR 0.0025 SAVE_CHECKPOINTS $SAVE_CHECKPOINTS OUTPUT_DIR $OUTPUT_DIR
        
 popd

From 648c70a92cae2b85c2de407d2549e358bb36c33d Mon Sep 17 00:00:00 2001
From: David Jurado <david.jurado@factored.ai>
Date: Tue, 3 Aug 2021 12:02:32 -0500
Subject: [PATCH 04/14] specify other parameter files location

---
 object_detection/README.md                        | 8 ++++++++
 object_detection/mlcube/workspace/parameters.yaml | 5 ++++-
 2 files changed, 12 insertions(+), 1 deletion(-)

diff --git a/object_detection/README.md b/object_detection/README.md
index e303bce26..4768a9161 100644
--- a/object_detection/README.md
+++ b/object_detection/README.md
@@ -30,7 +30,15 @@ The COCO dataset will be downloaded and extracted. Sizes of the dataset in each
 | Total                          | (After all tasks) | All            | ~41.7 GB |
 
 ### Tasks execution
+
+Parameters are defined at these files:
+
+* MLCube user parameters: mlcube/workspace/parameters.yaml
+* Project user parameters: pytorch/configs/e2e_mask_rcnn_R_50_FPN_1x.yaml
+* Project default parameters: pytorch/maskrcnn_benchmark/config/defaults.py
+
 ```
+
 # Download COCO dataset. Default path = /workspace/data
 python mlcube_cli.py run --task download_data --platform docker
 
diff --git a/object_detection/mlcube/workspace/parameters.yaml b/object_detection/mlcube/workspace/parameters.yaml
index 473864770..11d6035bf 100644
--- a/object_detection/mlcube/workspace/parameters.yaml
+++ b/object_detection/mlcube/workspace/parameters.yaml
@@ -1,2 +1,5 @@
 SAVE_CHECKPOINTS: "True" # Instead of False use empty value
-SOLVER_MAX_ITER: "100"
\ No newline at end of file
+SOLVER_MAX_ITER: "100"
+
+# Other user parameters are defined at ../../pytorch/configs/e2e_mask_rcnn_R_50_FPN_1x.yaml
+# Default parameters are defined at ../../pytorch/maskrcnn_benchmark/config/defaults.py
\ No newline at end of file

From 16f75fdaacd71009d7391f86583479502dcf4b45 Mon Sep 17 00:00:00 2001
From: David Jurado <david.jurado@factored.ai>
Date: Fri, 6 Aug 2021 10:07:12 -0500
Subject: [PATCH 05/14] Add support to override parameters at command line

---
 object_detection/mlcube/mlcube_cli.py | 22 +++++++++++++++++++---
 1 file changed, 19 insertions(+), 3 deletions(-)

diff --git a/object_detection/mlcube/mlcube_cli.py b/object_detection/mlcube/mlcube_cli.py
index 5df6a4bff..8c4a6643a 100644
--- a/object_detection/mlcube/mlcube_cli.py
+++ b/object_detection/mlcube/mlcube_cli.py
@@ -33,6 +33,21 @@ def load_config(mlcube_config_path: str, user_config_path: str) -> typing.Dict:
     return mlcube_config_data
 
 
+def override_extra_parameters(ctx, mlcube_config_data, task):
+    """Get extra paramters from context and override them on mlcube config data dict"""
+    for input_param in ctx.args:
+        input_key, input_value = input_param.split("=")
+        input_key = input_key.replace("--", "")
+        # Replace main container parameters
+        for key in mlcube_config_data["container"]:
+            if key==input_key:
+                mlcube_config_data["container"][key]=input_value
+        # Replace io paths in current task
+        for io in mlcube_config_data["tasks"][task]["io"]:
+            if io["name"]==input_key:
+                io["default"]=input_value
+
+
 @click.group(name='mlcube')
 def cli():
     pass
@@ -45,7 +60,8 @@ def cli():
 @click.option('--task', required=False, type=str, help='MLCube task name to run, default is `main`.')
 @click.option('--workspace', required=False, type=str, help='Workspace path, default is `workspace` within '
                                                             'MLCube folder')
-def run(mlcube: str, platform: str, task: str, workspace: str):
+@click.pass_context
+def run(ctx, mlcube: str, platform: str, task: str, workspace: str):
     mlcube_root = os.path.abspath(mlcube or os.getcwd())
     if os.path.isfile(mlcube_root):
         mlcube_root = os.path.dirname(mlcube_root)
@@ -61,10 +77,10 @@ def run(mlcube: str, platform: str, task: str, workspace: str):
         os.path.join(str(mlcube_root), 'mlcube.yaml'),
         os.path.join(os.path.expanduser("~"), '.mlcube.yaml')
     )
-
+    override_extra_parameters(ctx, mlcube_config_data, task)
     docker_runner = DockerRun(mlcube_config_data, root=mlcube_root, workspace=workspace, task=task)
     docker_runner.run()
 
 
 if __name__ == "__main__":
-    cli()
\ No newline at end of file
+    cli()

From efca6e008aa057cce169e1b251588fba441c271a Mon Sep 17 00:00:00 2001
From: David Jurado <david.jurado@factored.ai>
Date: Tue, 7 Sep 2021 20:47:04 -0500
Subject: [PATCH 06/14] Update to MLCube config v2.0

---
 object_detection/README.md            | 33 +++++-----
 object_detection/mlcube/mlcube.yaml   | 17 +++---
 object_detection/mlcube/mlcube_cli.py | 86 ---------------------------
 3 files changed, 26 insertions(+), 110 deletions(-)
 delete mode 100644 object_detection/mlcube/mlcube_cli.py

diff --git a/object_detection/README.md b/object_detection/README.md
index 4768a9161..4ddb4f37c 100644
--- a/object_detection/README.md
+++ b/object_detection/README.md
@@ -1,16 +1,15 @@
-## Current implementation
+# Benchmark execution with MLCube
 
-We'll be updating this section as we merge MLCube PRs and make new MLCube releases.
+## Project setup
 
-### Project setup
-```Python
+```bash
 # Create Python environment 
 virtualenv -p python3 ./env && source ./env/bin/activate
 
 # Install MLCube and MLCube docker runner from GitHub repository (normally, users will just run `pip install mlcube mlcube_docker`)
-git clone https://github.com/sergey-serebryakov/mlbox.git && cd mlbox && git checkout feature/configV2
-cd ./runners/mlcube_docker && export PYTHONPATH=$(pwd)
-cd ../../ && pip install -r mlcube/requirements.txt && pip install omegaconf && cd ../
+git clone https://github.com/mlcommons/mlcube && cd mlcube/mlcube
+python setup.py bdist_wheel  && pip install --force-reinstall ./dist/mlcube-* && cd ..
+cd ./runners/mlcube_docker && python setup.py bdist_wheel  && pip install --force-reinstall --no-deps ./dist/mlcube_docker-* && cd ../../..
 
 # Fetch the Object Detection workload
 git clone https://github.com/mlcommons/training && cd ./training
@@ -18,8 +17,7 @@ git fetch origin pull/501/head:feature/object_detection && git checkout feature/
 cd ./object_detection/mlcube
 ```
 
-### Dataset
-
+## Dataset
 
 The COCO dataset will be downloaded and extracted. Sizes of the dataset in each step:
 
@@ -29,7 +27,7 @@ The COCO dataset will be downloaded and extracted. Sizes of the dataset in each
 | Extract (Uncompressed dataset) | download_data     | Jpg/Json files | ~21.2 GB |
 | Total                          | (After all tasks) | All            | ~41.7 GB |
 
-### Tasks execution
+## Tasks execution
 
 Parameters are defined at these files:
 
@@ -37,11 +35,16 @@ Parameters are defined at these files:
 * Project user parameters: pytorch/configs/e2e_mask_rcnn_R_50_FPN_1x.yaml
 * Project default parameters: pytorch/maskrcnn_benchmark/config/defaults.py
 
-```
-
+```bash
 # Download COCO dataset. Default path = /workspace/data
-python mlcube_cli.py run --task download_data --platform docker
+mlcube run --task download_data
 
 # Run benchmark. Default paths = ./workspace/data
-python mlcube_cli.py run --task train --platform docker
-```
\ No newline at end of file
+mlcube run --task train
+```
+
+By default MLCube images use pull-type installation, so they should be available on docker hub. If not, try this:
+
+```bash
+mlcube run ... -Pdocker.build_strategy=auto
+```
diff --git a/object_detection/mlcube/mlcube.yaml b/object_detection/mlcube/mlcube.yaml
index a921adfec..288bc7d15 100644
--- a/object_detection/mlcube/mlcube.yaml
+++ b/object_detection/mlcube/mlcube.yaml
@@ -7,7 +7,7 @@ platform:
   # Edit this according to your system specs
   accelerator_count: 1
 
-container:
+docker:
   # Image name.
   image: mlcommons/object_detection:0.0.1
   # Docker build context relative to $MLCUBE_ROOT. Default is `build`.
@@ -18,15 +18,14 @@ container:
 tasks:
   # Download LibriSpeech dataset
   download_data:
-    io:
+    parameters:
       # Directory for uncompressed datasets. Total size is ~ 65G.
-      - {name: data_dir, type: directory, io: output, default: $WORKSPACE/data}
+      outputs: {data_dir: data/}
   train:
   # Train Object Detection model
-    io:
-      # see Download::data_dir, it takes the processes subfolder
-      - {name: data_dir, type: directory, io: input, default: $WORKSPACE/data}
+    parameters:
+      # data_dir: see Download::data_dir, it takes the processes subfolder
+      # parameters_file: Yaml file with training parameters.
+      inputs: {data_dir: data/, parameters_file: {type: file, default: parameters.yaml}}
       # Model output folder
-      - {name: output_dir, type: directory, io: output, default: $WORKSPACE/output}
-      # Yaml file with training parameters.
-      - {name: parameters_file, type: file, io: input, default: $WORKSPACE/parameters.yaml}
\ No newline at end of file
+      outputs: {output_dir: output/}
\ No newline at end of file
diff --git a/object_detection/mlcube/mlcube_cli.py b/object_detection/mlcube/mlcube_cli.py
deleted file mode 100644
index 8c4a6643a..000000000
--- a/object_detection/mlcube/mlcube_cli.py
+++ /dev/null
@@ -1,86 +0,0 @@
-"""
-This requires the MLCube 2.0 configuration
-"""
-import os
-import yaml
-import click
-import typing
-from mlcube_docker.docker_run import DockerRun
-
-
-def load_config(mlcube_config_path: str, user_config_path: str) -> typing.Dict:
-    """Returns dictionary containing MLCube configuration"""
-    # Load mlcube config data
-    try:
-        with open(mlcube_config_path) as stream:
-            mlcube_config_data = yaml.load(stream.read(), Loader=yaml.SafeLoader)
-    except IOError as exc:
-        # If file doesn't exist throw the exception:
-        # OSError: {PATH_TO}/mnist/mlcube.yaml: No such file or directory
-        raise IOError("%s: %s" % (mlcube_config_path, exc.strerror))
-
-    # Load user config data if file exists
-    if os.path.isfile(user_config_path):
-        with open(user_config_path) as stream:
-            user_config_data = yaml.load(stream.read(), Loader=yaml.SafeLoader)
-    else:
-        return mlcube_config_data
-
-    # Merge config data
-    tmp = mlcube_config_data['container']
-    mlcube_config_data['container'] = user_config_data['container']
-    mlcube_config_data['container'].update(tmp)
-    return mlcube_config_data
-
-
-def override_extra_parameters(ctx, mlcube_config_data, task):
-    """Get extra paramters from context and override them on mlcube config data dict"""
-    for input_param in ctx.args:
-        input_key, input_value = input_param.split("=")
-        input_key = input_key.replace("--", "")
-        # Replace main container parameters
-        for key in mlcube_config_data["container"]:
-            if key==input_key:
-                mlcube_config_data["container"][key]=input_value
-        # Replace io paths in current task
-        for io in mlcube_config_data["tasks"][task]["io"]:
-            if io["name"]==input_key:
-                io["default"]=input_value
-
-
-@click.group(name='mlcube')
-def cli():
-    pass
-
-
-@cli.command(name='run', help='Run MLCube ML task.',
-             context_settings=dict(ignore_unknown_options=True, allow_extra_args=True))
-@click.option('--mlcube', required=False, type=str, help='Path to MLCube directory, default is current.')
-@click.option('--platform', required=False, type=str, help='Platform to run MLCube, default is docker/podman.')
-@click.option('--task', required=False, type=str, help='MLCube task name to run, default is `main`.')
-@click.option('--workspace', required=False, type=str, help='Workspace path, default is `workspace` within '
-                                                            'MLCube folder')
-@click.pass_context
-def run(ctx, mlcube: str, platform: str, task: str, workspace: str):
-    mlcube_root = os.path.abspath(mlcube or os.getcwd())
-    if os.path.isfile(mlcube_root):
-        mlcube_root = os.path.dirname(mlcube_root)
-
-    platform = platform or 'docker'
-    if platform != 'docker':
-        raise ValueError(f"Only `docker` platform is supported")
-
-    task = task or 'main'
-    workspace = workspace or os.path.join(mlcube_root, 'workspace')
-
-    mlcube_config_data = load_config(
-        os.path.join(str(mlcube_root), 'mlcube.yaml'),
-        os.path.join(os.path.expanduser("~"), '.mlcube.yaml')
-    )
-    override_extra_parameters(ctx, mlcube_config_data, task)
-    docker_runner = DockerRun(mlcube_config_data, root=mlcube_root, workspace=workspace, task=task)
-    docker_runner.run()
-
-
-if __name__ == "__main__":
-    cli()

From 9395cfd931491aa722a810179e153e436f85a387 Mon Sep 17 00:00:00 2001
From: David Jurado <david.jurado@factored.ai>
Date: Fri, 22 Jul 2022 10:56:11 -0500
Subject: [PATCH 07/14] Update readme

---
 object_detection/README.md | 34 ++++++++++++++++++----------------
 1 file changed, 18 insertions(+), 16 deletions(-)

diff --git a/object_detection/README.md b/object_detection/README.md
index 4ddb4f37c..e8ac3b0a2 100644
--- a/object_detection/README.md
+++ b/object_detection/README.md
@@ -1,15 +1,13 @@
 # Benchmark execution with MLCube
 
-## Project setup
+## Current implementation
 
-```bash
-# Create Python environment 
-virtualenv -p python3 ./env && source ./env/bin/activate
+We'll be updating this section as we merge MLCube PRs and make new MLCube releases.
 
-# Install MLCube and MLCube docker runner from GitHub repository (normally, users will just run `pip install mlcube mlcube_docker`)
-git clone https://github.com/mlcommons/mlcube && cd mlcube/mlcube
-python setup.py bdist_wheel  && pip install --force-reinstall ./dist/mlcube-* && cd ..
-cd ./runners/mlcube_docker && python setup.py bdist_wheel  && pip install --force-reinstall --no-deps ./dist/mlcube_docker-* && cd ../../..
+### Project setup
+```Python
+# Create Python environment and install MLCube Docker runner 
+virtualenv -p python3 ./env && source ./env/bin/activate && pip install mlcube-docker
 
 # Fetch the Object Detection workload
 git clone https://github.com/mlcommons/training && cd ./training
@@ -17,7 +15,8 @@ git fetch origin pull/501/head:feature/object_detection && git checkout feature/
 cd ./object_detection/mlcube
 ```
 
-## Dataset
+### Dataset
+
 
 The COCO dataset will be downloaded and extracted. Sizes of the dataset in each step:
 
@@ -27,7 +26,7 @@ The COCO dataset will be downloaded and extracted. Sizes of the dataset in each
 | Extract (Uncompressed dataset) | download_data     | Jpg/Json files | ~21.2 GB |
 | Total                          | (After all tasks) | All            | ~41.7 GB |
 
-## Tasks execution
+### Tasks execution
 
 Parameters are defined at these files:
 
@@ -35,16 +34,19 @@ Parameters are defined at these files:
 * Project user parameters: pytorch/configs/e2e_mask_rcnn_R_50_FPN_1x.yaml
 * Project default parameters: pytorch/maskrcnn_benchmark/config/defaults.py
 
-```bash
+```
+
 # Download COCO dataset. Default path = /workspace/data
-mlcube run --task download_data
+python mlcube_cli.py run --task download_data --platform docker
 
 # Run benchmark. Default paths = ./workspace/data
-mlcube run --task train
+python mlcube_cli.py run --task train --platform docker
 ```
 
-By default MLCube images use pull-type installation, so they should be available on docker hub. If not, try this:
+Parameters defined at **mculbe/mlcube.yaml** could be overridden using: `--param=input`
+
+We are targeting pull-type installation, so MLCube images should be available on docker hub. If not, try this:
 
 ```bash
-mlcube run ... -Pdocker.build_strategy=auto
-```
+mlcube run ... -Pdocker.build_strategy=always
+```
\ No newline at end of file

From 3a9fe5f8e97b2cc45f9fae7edf362b5ee5e76365 Mon Sep 17 00:00:00 2001
From: David Jurado <david.jurado@factored.ai>
Date: Fri, 22 Mar 2024 15:56:13 -0500
Subject: [PATCH 08/14] Add MD5 verification

---
 object_detection/Dockerfile.mlcube   |  5 +++--
 object_detection/download_dataset.sh | 25 +++++++++++++++++++------
 2 files changed, 22 insertions(+), 8 deletions(-)

diff --git a/object_detection/Dockerfile.mlcube b/object_detection/Dockerfile.mlcube
index 5fe9dca29..43bd303a2 100644
--- a/object_detection/Dockerfile.mlcube
+++ b/object_detection/Dockerfile.mlcube
@@ -1,10 +1,11 @@
 FROM pytorch/pytorch:1.6.0-cuda10.1-cudnn7-devel
 
 RUN echo 'debconf debconf/frontend select Noninteractive' | debconf-set-selections
+RUN apt-key adv --fetch-keys https://developer.download.nvidia.com/compute/cuda/repos/ubuntu1804/x86_64/3bf863cc.pub
 
 # install basics
 RUN apt-get update -y \
- && apt-get install -y apt-utils curl unzip \
+ && apt-get install -y apt-utils curl unzip pv \
                        libglib2.0-0=2.56.1-2ubuntu1 \
                        libsm6=2:1.2.2-1 \
                        libxext6=2:1.3.3-1 \
@@ -26,7 +27,7 @@ WORKDIR /workspace
 
 COPY . .
 
-RUN chmod +x ./run_and_time.sh ./download_dataset.sh ./install.sh 
+RUN chmod +x ./run_and_time.sh ./download_dataset.sh
 
 ENV PYTHONPATH /workspace/pytorch
 
diff --git a/object_detection/download_dataset.sh b/object_detection/download_dataset.sh
index fcd857443..b1c0ded29 100755
--- a/object_detection/download_dataset.sh
+++ b/object_detection/download_dataset.sh
@@ -29,11 +29,24 @@ echo "Extracting train2017.zip:"
 n_files=`unzip -l  train2017.zip| grep .jpg | wc -l`
 unzip train2017.zip  | { I=-1; while read; do printf "Progress: $((++I*100/$n_files))%%\r"; done; echo ""; }
 
-# TBD: MD5 verification
-# $md5sum *.zip *.tgz
-#f4bbac642086de4f52a3fdda2de5fa2c  annotations_trainval2017.zip
-#cced6f7f71b7629ddf16f17bbcfab6b2  train2017.zip
-#442b8da7639aecaf257c1dceb8ba8c80  val2017.zip
-#2d2b9d2283adb5e3b8d25eec88e65064  coco_annotations_minival.tgz
+# MD5 verification
+echo "Running MD5 verification ... this might take a while"
+checkMD5 () {
+    if [ $(pv -f $1| md5sum | cut -d' ' -f1) = $2 ];
+    then
+        echo "$1 MD5 is valid"
+    else 
+        echo "*ERROR* $1 MD5 is NOT valid"
+    fi
+}
+
+echo "validating annotations_trainval2017.zip:"
+checkMD5 "annotations_trainval2017.zip" "f4bbac642086de4f52a3fdda2de5fa2c"
+echo "validating coco_annotations_minival.tgz:"
+checkMD5 "coco_annotations_minival.tgz" "2d2b9d2283adb5e3b8d25eec88e65064"
+echo "validating val2017.zip:"
+checkMD5 "val2017.zip" "442b8da7639aecaf257c1dceb8ba8c80"
+echo "validating train2017.zip:"
+checkMD5 "train2017.zip" "cced6f7f71b7629ddf16f17bbcfab6b2"
 
 popd

From ef001bb64771ba1725c143c4dc1be13b59fca2c9 Mon Sep 17 00:00:00 2001
From: David Jurado <david.jurado@factored.ai>
Date: Fri, 22 Mar 2024 16:01:02 -0500
Subject: [PATCH 09/14] Fix SAVE_CHECKPOINTS flag

---
 object_detection/mlcube/workspace/parameters.yaml | 2 +-
 1 file changed, 1 insertion(+), 1 deletion(-)

diff --git a/object_detection/mlcube/workspace/parameters.yaml b/object_detection/mlcube/workspace/parameters.yaml
index 11d6035bf..f90a6c9fd 100644
--- a/object_detection/mlcube/workspace/parameters.yaml
+++ b/object_detection/mlcube/workspace/parameters.yaml
@@ -1,4 +1,4 @@
-SAVE_CHECKPOINTS: "True" # Instead of False use empty value
+SAVE_CHECKPOINTS: "" # Instead of False use empty value
 SOLVER_MAX_ITER: "100"
 
 # Other user parameters are defined at ../../pytorch/configs/e2e_mask_rcnn_R_50_FPN_1x.yaml

From 711ecaa440aabc9c8dceef821efb289ac90b6c09 Mon Sep 17 00:00:00 2001
From: David Jurado <david.jurado@factored.ai>
Date: Fri, 22 Mar 2024 16:13:15 -0500
Subject: [PATCH 10/14] Fix MLCube readme

---
 object_detection/README_OLD.md    | 95 -------------------------------
 object_detection/mlcube/README.md |  6 +-
 2 files changed, 3 insertions(+), 98 deletions(-)
 delete mode 100644 object_detection/README_OLD.md

diff --git a/object_detection/README_OLD.md b/object_detection/README_OLD.md
deleted file mode 100644
index 6895735de..000000000
--- a/object_detection/README_OLD.md
+++ /dev/null
@@ -1,95 +0,0 @@
-# 1. Problem
-Object detection and segmentation. Metrics are mask and box mAP.
-
-# 2. Directions
-
-### Steps to configure machine
-
-1. Checkout the MLPerf repository
-```
-mkdir -p mlperf
-cd mlperf
-git clone https://github.com/mlperf/training.git
-```
-2. Install CUDA and Docker
-```
-source training/install_cuda_docker.sh
-```
-3. Build the docker image for the object detection task
-```
-cd training/object_detection/
-nvidia-docker build . -t mlperf/object_detection
-```
-
-4. Run docker container and install code
-```
-nvidia-docker run -v .:/workspace -t -i --rm --ipc=host mlperf/object_detection \
-    "cd mlperf/training/object_detection && ./install.sh"
-```
-Now exit the docker container (Ctrl-D) to get back to your host.
-
-### Steps to download data
-```
-# From training/object_detection/
-source download_dataset.sh
-```
-
-### Steps to run benchmark.
-```
-nvidia-docker run -v .:/workspace -t -i --rm --ipc=host mlperf/object_detection \
-    "cd mlperf/training/object_detection && ./run_and_time.sh"
-```
-
-# 3. Dataset/Environment
-### Publication/Attribution
-Microsoft COCO: Common Objects in Context
-
-### Data preprocessing
-Only horizontal flips are allowed.
-
-### Training and test data separation
-As provided by MS-COCO (2017 version).
-
-### Training data order
-Randomly.
-
-### Test data order
-Any order.
-
-# 4. Model
-### Publication/Attribution
-He, Kaiming, et al. "Mask r-cnn." Computer Vision (ICCV), 2017 IEEE International Conference on.
-IEEE, 2017.
-
-We use a version of Mask R-CNN with a ResNet50 backbone.
-
-### List of layers
-Running the timing script will display a list of layers.
-
-### Weight and bias initialization
-The ResNet50 base must be loaded from the provided weights. They may be quantized.
-
-### Loss function
-Multi-task loss (classification, box, mask). Described in the Mask R-CNN paper.
-
-Classification: Smooth L1 loss
-
-Box: Log loss for true class.
-
-Mask: per-pixel sigmoid, average binary cross-entropy loss.
-
-### Optimizer
-Momentum SGD. Weight decay of 0.0001, momentum of 0.9.
-
-# 5. Quality
-### Quality metric
-As Mask R-CNN can provide both boxes and masks, we evaluate on both box and mask mAP.
-
-### Quality target
-Box mAP of 0.377, mask mAP of 0.339
-
-### Evaluation frequency
-Once per epoch, 118k.
-
-### Evaluation thoroughness
-Evaluate over the entire validation set. Use the official COCO API to compute mAP.
diff --git a/object_detection/mlcube/README.md b/object_detection/mlcube/README.md
index e8ac3b0a2..f1e17ba11 100644
--- a/object_detection/mlcube/README.md
+++ b/object_detection/mlcube/README.md
@@ -37,10 +37,10 @@ Parameters are defined at these files:
 ```
 
 # Download COCO dataset. Default path = /workspace/data
-python mlcube_cli.py run --task download_data --platform docker
+mlcube run --task download_data --platform docker
 
 # Run benchmark. Default paths = ./workspace/data
-python mlcube_cli.py run --task train --platform docker
+mlcube run --task train --platform docker
 ```
 
 Parameters defined at **mculbe/mlcube.yaml** could be overridden using: `--param=input`
@@ -49,4 +49,4 @@ We are targeting pull-type installation, so MLCube images should be available on
 
 ```bash
 mlcube run ... -Pdocker.build_strategy=always
-```
\ No newline at end of file
+```

From 0db618f6f2cba014845f656093212d06fca48893 Mon Sep 17 00:00:00 2001
From: David Jurado <david.jurado@factored.ai>
Date: Fri, 19 Apr 2024 14:19:54 -0500
Subject: [PATCH 11/14] Add demo tasks

---
 object_detection/Dockerfile.mlcube   |  4 +--
 object_detection/download_dataset.sh | 13 +++++++-
 object_detection/download_demo.sh    | 39 ++++++++++++++++++++++++
 object_detection/mlcube/.gitignore   |  2 ++
 object_detection/mlcube/README.md    | 31 +++++++++++++++-----
 object_detection/mlcube/mlcube.yaml  | 15 +++++++++-
 object_detection/run_and_time.sh     | 36 ++++++++++++++++-------
 object_detection/run_demo.sh         | 44 ++++++++++++++++++++++++++++
 8 files changed, 161 insertions(+), 23 deletions(-)
 create mode 100755 object_detection/download_demo.sh
 create mode 100644 object_detection/mlcube/.gitignore
 create mode 100755 object_detection/run_demo.sh

diff --git a/object_detection/Dockerfile.mlcube b/object_detection/Dockerfile.mlcube
index 43bd303a2..5b802e3f9 100644
--- a/object_detection/Dockerfile.mlcube
+++ b/object_detection/Dockerfile.mlcube
@@ -29,6 +29,4 @@ COPY . .
 
 RUN chmod +x ./run_and_time.sh ./download_dataset.sh
 
-ENV PYTHONPATH /workspace/pytorch
-
-ENTRYPOINT ["python", "/workspace/mlcube.py"]
\ No newline at end of file
+ENV PYTHONPATH /workspace/pytorch
\ No newline at end of file
diff --git a/object_detection/download_dataset.sh b/object_detection/download_dataset.sh
index b1c0ded29..8203c2c92 100755
--- a/object_detection/download_dataset.sh
+++ b/object_detection/download_dataset.sh
@@ -1,7 +1,18 @@
 #!/bin/bash
+: "${DATA_ROOT_DIR:=./pytorch/datasets/coco}"
+
+while [ $# -gt 0 ]; do
+  case "$1" in
+  --data_dir=*)
+    DATA_ROOT_DIR="${1#*=}"
+    ;;
+  *) ;;
+  esac
+  shift
+done
+
 
 # Get COCO 2017 data sets
-DATA_ROOT_DIR="${DATA_ROOT_DIR:-./pytorch/datasets/coco}"
 echo "Downloaading to folder: $DATA_ROOT_DIR"
 mkdir -p $DATA_ROOT_DIR
 pushd $DATA_ROOT_DIR
diff --git a/object_detection/download_demo.sh b/object_detection/download_demo.sh
new file mode 100755
index 000000000..f2d9896f9
--- /dev/null
+++ b/object_detection/download_demo.sh
@@ -0,0 +1,39 @@
+#!/bin/bash
+: "${DATA_ROOT_DIR:=./pytorch/datasets/coco}"
+
+while [ $# -gt 0 ]; do
+  case "$1" in
+  --data_dir=*)
+    DATA_ROOT_DIR="${1#*=}"
+    ;;
+  *) ;;
+  esac
+  shift
+done
+
+
+echo "Downloaading demo to folder: $DATA_ROOT_DIR"
+mkdir -p $DATA_ROOT_DIR
+pushd $DATA_ROOT_DIR
+
+echo "Downloading annotations_trainval2017.zip:"
+curl -O https://storage.googleapis.com/mlperf_training_demo/object_detection/demo_data.zip
+echo "Extracting demo_data.zip:"
+unzip -o -q demo_data.zip
+echo "Done!"
+
+# MD5 verification
+echo "Running MD5 verification ..."
+checkMD5 () {
+    if [ $(pv -f $1| md5sum | cut -d' ' -f1) = $2 ];
+    then
+        echo "$1 MD5 is valid"
+    else 
+        echo "*ERROR* $1 MD5 is NOT valid"
+    fi
+}
+
+echo "validating demo_data.zip:"
+checkMD5 "demo_data.zip" "1b50202a21b0d8c3235d0a6f39b6f40c"
+
+popd
diff --git a/object_detection/mlcube/.gitignore b/object_detection/mlcube/.gitignore
new file mode 100644
index 000000000..98541c90f
--- /dev/null
+++ b/object_detection/mlcube/.gitignore
@@ -0,0 +1,2 @@
+workspace/
+!workspace/parameter.yaml
\ No newline at end of file
diff --git a/object_detection/mlcube/README.md b/object_detection/mlcube/README.md
index f1e17ba11..8d77a2925 100644
--- a/object_detection/mlcube/README.md
+++ b/object_detection/mlcube/README.md
@@ -1,9 +1,5 @@
 # Benchmark execution with MLCube
 
-## Current implementation
-
-We'll be updating this section as we merge MLCube PRs and make new MLCube releases.
-
 ### Project setup
 ```Python
 # Create Python environment and install MLCube Docker runner 
@@ -34,15 +30,34 @@ Parameters are defined at these files:
 * Project user parameters: pytorch/configs/e2e_mask_rcnn_R_50_FPN_1x.yaml
 * Project default parameters: pytorch/maskrcnn_benchmark/config/defaults.py
 
-```
-
+```bash
 # Download COCO dataset. Default path = /workspace/data
-mlcube run --task download_data --platform docker
+mlcube run --task=download_data -Pdocker.build_strategy=always
 
 # Run benchmark. Default paths = ./workspace/data
-mlcube run --task train --platform docker
+mlcube run --task=train -Pdocker.build_strategy=always
+```
+
+### Demo execution
+
+These tasks will use a demo dataset (39M) to execute a faster training workload for a quick demo (~12 min):
+
+```bash
+# Download subsampled dataset. Default path = /workspace/demo
+mlcube run --task=download_data -Pdocker.build_strategy=always
+
+# Run benchmark. Default paths = ./workspace/demo and ./workspace/demo_output
+mlcube run --task=demo -Pdocker.build_strategy=always
+```
+
+It's also possible to execute the two tasks in one single instruction:
+
+```bash
+mlcube run --task=download_demo,demo -Pdocker.build_strategy=always
 ```
 
+### Aditonal options
+
 Parameters defined at **mculbe/mlcube.yaml** could be overridden using: `--param=input`
 
 We are targeting pull-type installation, so MLCube images should be available on docker hub. If not, try this:
diff --git a/object_detection/mlcube/mlcube.yaml b/object_detection/mlcube/mlcube.yaml
index 288bc7d15..b0c693d58 100644
--- a/object_detection/mlcube/mlcube.yaml
+++ b/object_detection/mlcube/mlcube.yaml
@@ -14,18 +14,31 @@ docker:
   build_context: "../"
   # Docker file name within docker build context, default is `Dockerfile`.
   build_file: "Dockerfile.mlcube"
+  gpu_args: "--gpus=all --shm-size=1g"
 
 tasks:
   # Download LibriSpeech dataset
   download_data:
+    entrypoint: ./download_dataset.sh -a
     parameters:
       # Directory for uncompressed datasets. Total size is ~ 65G.
       outputs: {data_dir: data/}
   train:
   # Train Object Detection model
+    entrypoint: ./run_and_time.sh -a
     parameters:
       # data_dir: see Download::data_dir, it takes the processes subfolder
       # parameters_file: Yaml file with training parameters.
       inputs: {data_dir: data/, parameters_file: {type: file, default: parameters.yaml}}
       # Model output folder
-      outputs: {output_dir: output/}
\ No newline at end of file
+      outputs: {output_dir: output/}
+  download_demo:
+    entrypoint: ./download_demo.sh -a
+    parameters:
+      # Directory for uncompressed datasets. Total size is ~ 65G.
+      outputs: {data_dir: demo/}
+  demo:
+    entrypoint: ./run_demo.sh -a
+    parameters:
+      inputs: {data_dir: demo/, parameters_file: {type: file, default: parameters.yaml}}
+      outputs: {output_dir: demo_output/}
\ No newline at end of file
diff --git a/object_detection/run_and_time.sh b/object_detection/run_and_time.sh
index f7709d19e..69cd434c6 100755
--- a/object_detection/run_and_time.sh
+++ b/object_detection/run_and_time.sh
@@ -1,11 +1,29 @@
 #!/bin/bash
+# Runs benchmark and reports time to convergence
+pushd pytorch
+python setup.py clean build develop --user
+
+: "${DATA_DIR:=pytorch/datasets/coco}"
+: "${OUTPUT_DIR:=pytorch/output}"
 
+while [ $# -gt 0 ]; do
+  case "$1" in
+  --data_dir=*)
+    DATA_DIR="${1#*=}"
+    ;;
+  --output_dir=*)
+    OUTPUT_DIR="${1#*=}"
+    ;;
+  *) ;;
+  esac
+  shift
+done
 
-DATA_DIR="${DATA_DIR:-pytorch/datasets/coco}"
-DATA_ROOT_TARGET="pytorch/datasets/coco"
-OUTPUT_DIR="${OUTPUT_DIR:-}"
+echo "DATA_DIR"
+echo $DATA_DIR
 
-SAVE_CHECKPOINTS="${SAVE_CHECKPOINTS:-}"
+DATA_ROOT_TARGET="datasets/coco"
+SAVE_CHECKPOINTS="${SAVE_CHECKPOINTS:-False}"
 SOLVER_MAX_ITER="${SOLVER_MAX_ITER:-40000}"
 
 #Link input data paths
@@ -15,15 +33,13 @@ if [ "$DATA_DIR" != "$DATA_ROOT_TARGET" ]; then
        ln -s $DATA_DIR/train2017 $DATA_ROOT_TARGET/train2017
        ln -s $DATA_DIR/test2017 $DATA_ROOT_TARGET/test2017
        ln -s $DATA_DIR/val2017 $DATA_ROOT_TARGET/val2017
-       echo $DATA_ROOT_TARGET
-       ls -lah $DATA_ROOT_TARGET
 fi
 
-# Runs benchmark and reports time to convergence
-pushd pytorch
+pwd
+
 # Single GPU training
 time python tools/train_mlperf.py --config-file "configs/e2e_mask_rcnn_R_50_FPN_1x.yaml" \
-       SOLVER.IMS_PER_BATCH 2 TEST.IMS_PER_BATCH 1 SOLVER.MAX_ITER 720000 SOLVER.STEPS "(480000, 640000)" \
-       SOLVER.MAX_ITER $SOLVER_MAX_ITER SOLVER.BASE_LR 0.0025 SAVE_CHECKPOINTS $SAVE_CHECKPOINTS OUTPUT_DIR $OUTPUT_DIR
+       SOLVER.IMS_PER_BATCH 2 TEST.IMS_PER_BATCH 1 SOLVER.MAX_ITER 720000 \
+       SOLVER.STEPS "(480000, 640000)" SOLVER.BASE_LR 0.0025 OUTPUT_DIR $OUTPUT_DIR
        
 popd
diff --git a/object_detection/run_demo.sh b/object_detection/run_demo.sh
new file mode 100755
index 000000000..072895ca6
--- /dev/null
+++ b/object_detection/run_demo.sh
@@ -0,0 +1,44 @@
+#!/bin/bash
+# Runs benchmark and reports time to convergence
+pushd pytorch
+python setup.py clean build develop --user
+
+: "${DATA_DIR:=pytorch/datasets/coco}"
+: "${OUTPUT_DIR:=pytorch/output}"
+
+while [ $# -gt 0 ]; do
+  case "$1" in
+  --data_dir=*)
+    DATA_DIR="${1#*=}"
+    ;;
+  --output_dir=*)
+    OUTPUT_DIR="${1#*=}"
+    ;;
+  *) ;;
+  esac
+  shift
+done
+
+echo "DATA_DIR"
+echo $DATA_DIR
+
+DATA_ROOT_TARGET="datasets/coco"
+SAVE_CHECKPOINTS="${SAVE_CHECKPOINTS:-False}"
+SOLVER_MAX_ITER="${SOLVER_MAX_ITER:-40000}"
+
+#Link input data paths
+if [ "$DATA_DIR" != "$DATA_ROOT_TARGET" ]; then
+       mkdir -p $DATA_ROOT_TARGET
+       ln -s $DATA_DIR/annotations $DATA_ROOT_TARGET/annotations
+       ln -s $DATA_DIR/train2017 $DATA_ROOT_TARGET/train2017
+       ln -s $DATA_DIR/val2017 $DATA_ROOT_TARGET/val2017
+fi
+
+pwd
+
+# Single GPU training
+time python tools/train_mlperf.py --config-file "configs/e2e_mask_rcnn_R_50_FPN_1x.yaml" \
+       SOLVER.IMS_PER_BATCH 2 TEST.IMS_PER_BATCH 1 SOLVER.MAX_ITER 720 \
+       SOLVER.STEPS "(480, 640)" SOLVER.BASE_LR 0.0025 OUTPUT_DIR $OUTPUT_DIR
+       
+popd

From 573f7823eb0ad66a9545902c1d6f101b5486f7d9 Mon Sep 17 00:00:00 2001
From: David Jurado <david.jurado@factored.ai>
Date: Fri, 26 Apr 2024 10:59:42 -0500
Subject: [PATCH 12/14] update scripts

---
 object_detection/README.md           | 27 ++++++++++-
 object_detection/download_dataset.sh | 38 ++++++++++------
 object_detection/download_demo.sh    | 14 +++---
 object_detection/mlcube.py           | 67 ----------------------------
 object_detection/run_and_time.sh     | 16 +++----
 object_detection/run_demo.sh         | 14 +++---
 6 files changed, 71 insertions(+), 105 deletions(-)
 delete mode 100644 object_detection/mlcube.py

diff --git a/object_detection/README.md b/object_detection/README.md
index 2761dad4d..2754eec03 100644
--- a/object_detection/README.md
+++ b/object_detection/README.md
@@ -1,4 +1,5 @@
 # 1. Problem
+
 Object detection and segmentation. Metrics are mask and box mAP.
 
 # 2. Directions
@@ -6,63 +7,81 @@ Object detection and segmentation. Metrics are mask and box mAP.
 ### Steps to configure machine
 
 1. Checkout the MLPerf repository
+
 ```
 mkdir -p mlperf
 cd mlperf
 git clone https://github.com/mlperf/training.git
 ```
+
 2. Install CUDA and Docker
+
 ```
 source training/install_cuda_docker.sh
 ```
+
 3. Build the docker image for the object detection task
+
 ```
 cd training/object_detection/
 nvidia-docker build . -t mlperf/object_detection
 ```
 
 ### Steps to download data
+
 ```
 # From training/object_detection/
 source download_dataset.sh
 ```
 
-### Steps to run benchmark.
+### Steps to run benchmark
+
 ```
 nvidia-docker run -v .:/workspace -t -i --rm --ipc=host mlperf/object_detection \
     "cd mlperf/training/object_detection && ./run_and_time.sh"
 ```
 
 # 3. Dataset/Environment
+
 ### Publication/Attribution
+
 Microsoft COCO: Common Objects in Context
 
 ### Data preprocessing
+
 Only horizontal flips are allowed.
 
 ### Training and test data separation
+
 As provided by MS-COCO (2017 version).
 
 ### Training data order
+
 Randomly.
 
 ### Test data order
+
 Any order.
 
 # 4. Model
+
 ### Publication/Attribution
+
 He, Kaiming, et al. "Mask r-cnn." Computer Vision (ICCV), 2017 IEEE International Conference on.
 IEEE, 2017.
 
 We use a version of Mask R-CNN with a ResNet-50 backbone.
 
 ### List of layers
+
 Running the timing script will display a list of layers.
 
 ### Weight and bias initialization
+
 The ResNet-50 base must be loaded from the provided weights. They may be quantized.
 
 ### Loss function
+
 Multi-task loss (classification, box, mask). Described in the Mask R-CNN paper.
 
 Classification: Smooth L1 loss
@@ -72,17 +91,23 @@ Box: Log loss for true class.
 Mask: per-pixel sigmoid, average binary cross-entropy loss.
 
 ### Optimizer
+
 Momentum SGD. Weight decay of 0.0001, momentum of 0.9.
 
 # 5. Quality
+
 ### Quality metric
+
 As Mask R-CNN can provide both boxes and masks, we evaluate on both box and mask mAP.
 
 ### Quality target
+
 Box mAP of 0.377, mask mAP of 0.339
 
 ### Evaluation frequency
+
 Once per epoch, 118k.
 
 ### Evaluation thoroughness
+
 Evaluate over the entire validation set. Use the official COCO API to compute mAP.
diff --git a/object_detection/download_dataset.sh b/object_detection/download_dataset.sh
index 8203c2c92..8b5c68d71 100755
--- a/object_detection/download_dataset.sh
+++ b/object_detection/download_dataset.sh
@@ -11,7 +11,6 @@ while [ $# -gt 0 ]; do
   shift
 done
 
-
 # Get COCO 2017 data sets
 echo "Downloaading to folder: $DATA_ROOT_DIR"
 mkdir -p $DATA_ROOT_DIR
@@ -25,30 +24,41 @@ tar -xzf coco_annotations_minival.tgz &>/dev/null
 echo "Downloading annotations_trainval2017.zip:"
 curl -O http://images.cocodataset.org/annotations/annotations_trainval2017.zip
 echo "Extracting annotations_trainval2017.zip:"
-n_files=`unzip -l  annotations_trainval2017.zip| grep .json | wc -l`
-unzip annotations_trainval2017.zip  | { I=-1; while read; do printf "Progress: $((++I*100/$n_files))%%\r"; done; echo ""; }
+n_files=$(unzip -l annotations_trainval2017.zip | grep .json | wc -l)
+unzip annotations_trainval2017.zip | {
+  I=-1
+  while read; do printf "Progress: $((++I * 100 / $n_files))%%\r"; done
+  echo ""
+}
 
 echo "Downloading val2017.zip:"
 curl -O http://images.cocodataset.org/zips/val2017.zip
 echo "Extracting val2017.zip:"
-n_files=`unzip -l  val2017.zip| grep .jpg | wc -l`
-unzip val2017.zip  | { I=-1; while read; do printf "Progress: $((++I*100/$n_files))%%\r"; done; echo ""; }
+n_files=$(unzip -l val2017.zip | grep .jpg | wc -l)
+unzip val2017.zip | {
+  I=-1
+  while read; do printf "Progress: $((++I * 100 / $n_files))%%\r"; done
+  echo ""
+}
 
 echo "Downloading train2017.zip:"
 curl -O http://images.cocodataset.org/zips/train2017.zip
 echo "Extracting train2017.zip:"
-n_files=`unzip -l  train2017.zip| grep .jpg | wc -l`
-unzip train2017.zip  | { I=-1; while read; do printf "Progress: $((++I*100/$n_files))%%\r"; done; echo ""; }
+n_files=$(unzip -l train2017.zip | grep .jpg | wc -l)
+unzip train2017.zip | {
+  I=-1
+  while read; do printf "Progress: $((++I * 100 / $n_files))%%\r"; done
+  echo ""
+}
 
 # MD5 verification
 echo "Running MD5 verification ... this might take a while"
-checkMD5 () {
-    if [ $(pv -f $1| md5sum | cut -d' ' -f1) = $2 ];
-    then
-        echo "$1 MD5 is valid"
-    else 
-        echo "*ERROR* $1 MD5 is NOT valid"
-    fi
+checkMD5() {
+  if [ $(pv -f $1 | md5sum | cut -d' ' -f1) = $2 ]; then
+    echo "$1 MD5 is valid"
+  else
+    echo "*ERROR* $1 MD5 is NOT valid"
+  fi
 }
 
 echo "validating annotations_trainval2017.zip:"
diff --git a/object_detection/download_demo.sh b/object_detection/download_demo.sh
index f2d9896f9..b2266604d 100755
--- a/object_detection/download_demo.sh
+++ b/object_detection/download_demo.sh
@@ -11,7 +11,6 @@ while [ $# -gt 0 ]; do
   shift
 done
 
-
 echo "Downloaading demo to folder: $DATA_ROOT_DIR"
 mkdir -p $DATA_ROOT_DIR
 pushd $DATA_ROOT_DIR
@@ -24,13 +23,12 @@ echo "Done!"
 
 # MD5 verification
 echo "Running MD5 verification ..."
-checkMD5 () {
-    if [ $(pv -f $1| md5sum | cut -d' ' -f1) = $2 ];
-    then
-        echo "$1 MD5 is valid"
-    else 
-        echo "*ERROR* $1 MD5 is NOT valid"
-    fi
+checkMD5() {
+  if [ $(pv -f $1 | md5sum | cut -d' ' -f1) = $2 ]; then
+    echo "$1 MD5 is valid"
+  else
+    echo "*ERROR* $1 MD5 is NOT valid"
+  fi
 }
 
 echo "validating demo_data.zip:"
diff --git a/object_detection/mlcube.py b/object_detection/mlcube.py
deleted file mode 100644
index b3f8c2d03..000000000
--- a/object_detection/mlcube.py
+++ /dev/null
@@ -1,67 +0,0 @@
-"""MLCube handler file"""
-import os
-import yaml
-import typer
-import shutil
-import subprocess
-from pathlib import Path
-
-
-app = typer.Typer()
-
-class DownloadDataTask(object):
-    """Download task Class
-    It defines the environment variables:
-        DATA_ROOT_DIR: Directory path to download the dataset
-    Then executes the download script"""
-    @staticmethod
-    def run(data_dir: str) -> None:
-
-        env = os.environ.copy()
-        env.update({
-            'DATA_ROOT_DIR': data_dir,
-        })
-
-        process = subprocess.Popen("./download_dataset.sh", cwd=".", env=env)
-        process.wait()
-
-class TrainTask(object):
-    """Preprocess dataset task Class
-    It defines the environment variables:
-        DATA_DIR: Dataset directory path
-        OUTPUT_DIR: Directory path where model will be saved
-        All other parameters defined in parameters_file
-    Then executes the benchmark script"""
-    @staticmethod
-    def run(data_dir: str, output_dir: str, parameters_file: str) -> None:
-
-        process = subprocess.Popen("./install.sh", cwd=".")
-        process.wait()
-
-        with open(parameters_file, 'r') as stream:
-            parameters = yaml.safe_load(stream)
-
-        env = os.environ.copy()
-        env.update({
-            'DATA_DIR': data_dir,
-            'OUTPUT_DIR': output_dir,
-        })
-
-        if parameters is not None:
-            env.update(parameters)
-
-        process = subprocess.Popen("./run_and_time.sh", cwd=".", env=env)
-        process.wait()
-
-@app.command("download_data")
-def download_data(data_dir: str = typer.Option(..., '--data_dir')):
-    DownloadDataTask.run(data_dir)
-
-@app.command("train")
-def train(data_dir: str = typer.Option(..., '--data_dir'),
-          output_dir: str = typer.Option(..., '--output_dir'),
-          parameters_file: str = typer.Option(..., '--parameters_file')):
-    TrainTask.run(data_dir, output_dir, parameters_file)
-
-if __name__ == '__main__':
-    app()
\ No newline at end of file
diff --git a/object_detection/run_and_time.sh b/object_detection/run_and_time.sh
index 69cd434c6..e26bc9125 100755
--- a/object_detection/run_and_time.sh
+++ b/object_detection/run_and_time.sh
@@ -28,18 +28,18 @@ SOLVER_MAX_ITER="${SOLVER_MAX_ITER:-40000}"
 
 #Link input data paths
 if [ "$DATA_DIR" != "$DATA_ROOT_TARGET" ]; then
-       mkdir -p $DATA_ROOT_TARGET
-       ln -s $DATA_DIR/annotations $DATA_ROOT_TARGET/annotations
-       ln -s $DATA_DIR/train2017 $DATA_ROOT_TARGET/train2017
-       ln -s $DATA_DIR/test2017 $DATA_ROOT_TARGET/test2017
-       ln -s $DATA_DIR/val2017 $DATA_ROOT_TARGET/val2017
+  mkdir -p $DATA_ROOT_TARGET
+  ln -s $DATA_DIR/annotations $DATA_ROOT_TARGET/annotations
+  ln -s $DATA_DIR/train2017 $DATA_ROOT_TARGET/train2017
+  ln -s $DATA_DIR/test2017 $DATA_ROOT_TARGET/test2017
+  ln -s $DATA_DIR/val2017 $DATA_ROOT_TARGET/val2017
 fi
 
 pwd
 
 # Single GPU training
 time python tools/train_mlperf.py --config-file "configs/e2e_mask_rcnn_R_50_FPN_1x.yaml" \
-       SOLVER.IMS_PER_BATCH 2 TEST.IMS_PER_BATCH 1 SOLVER.MAX_ITER 720000 \
-       SOLVER.STEPS "(480000, 640000)" SOLVER.BASE_LR 0.0025 OUTPUT_DIR $OUTPUT_DIR
-       
+  SOLVER.IMS_PER_BATCH 2 TEST.IMS_PER_BATCH 1 SOLVER.MAX_ITER 720000 \
+  SOLVER.STEPS "(480000, 640000)" SOLVER.BASE_LR 0.0025 OUTPUT_DIR $OUTPUT_DIR
+
 popd
diff --git a/object_detection/run_demo.sh b/object_detection/run_demo.sh
index 072895ca6..efb1c0c98 100755
--- a/object_detection/run_demo.sh
+++ b/object_detection/run_demo.sh
@@ -28,17 +28,17 @@ SOLVER_MAX_ITER="${SOLVER_MAX_ITER:-40000}"
 
 #Link input data paths
 if [ "$DATA_DIR" != "$DATA_ROOT_TARGET" ]; then
-       mkdir -p $DATA_ROOT_TARGET
-       ln -s $DATA_DIR/annotations $DATA_ROOT_TARGET/annotations
-       ln -s $DATA_DIR/train2017 $DATA_ROOT_TARGET/train2017
-       ln -s $DATA_DIR/val2017 $DATA_ROOT_TARGET/val2017
+  mkdir -p $DATA_ROOT_TARGET
+  ln -s $DATA_DIR/annotations $DATA_ROOT_TARGET/annotations
+  ln -s $DATA_DIR/train2017 $DATA_ROOT_TARGET/train2017
+  ln -s $DATA_DIR/val2017 $DATA_ROOT_TARGET/val2017
 fi
 
 pwd
 
 # Single GPU training
 time python tools/train_mlperf.py --config-file "configs/e2e_mask_rcnn_R_50_FPN_1x.yaml" \
-       SOLVER.IMS_PER_BATCH 2 TEST.IMS_PER_BATCH 1 SOLVER.MAX_ITER 720 \
-       SOLVER.STEPS "(480, 640)" SOLVER.BASE_LR 0.0025 OUTPUT_DIR $OUTPUT_DIR
-       
+  SOLVER.IMS_PER_BATCH 2 TEST.IMS_PER_BATCH 1 SOLVER.MAX_ITER 720 \
+  SOLVER.STEPS "(480, 640)" SOLVER.BASE_LR 0.0025 OUTPUT_DIR $OUTPUT_DIR
+
 popd

From 9f9ab233714cba9f8f570cdcb3e283f8a1a82dea Mon Sep 17 00:00:00 2001
From: David Jurado <david.jurado@factored.ai>
Date: Fri, 26 Jul 2024 09:43:41 -0500
Subject: [PATCH 13/14] update demo download link

---
 object_detection/download_demo.sh | 5 +++--
 1 file changed, 3 insertions(+), 2 deletions(-)

diff --git a/object_detection/download_demo.sh b/object_detection/download_demo.sh
index b2266604d..4a0fc10ac 100755
--- a/object_detection/download_demo.sh
+++ b/object_detection/download_demo.sh
@@ -16,9 +16,10 @@ mkdir -p $DATA_ROOT_DIR
 pushd $DATA_ROOT_DIR
 
 echo "Downloading annotations_trainval2017.zip:"
-curl -O https://storage.googleapis.com/mlperf_training_demo/object_detection/demo_data.zip
+curl -O https://mlcube.mlcommons-storage.org/minibenchmarks/object_detection.zip
 echo "Extracting demo_data.zip:"
-unzip -o -q demo_data.zip
+unzip -o -q object_detection.zip
+rm object_detection.zip
 echo "Done!"
 
 # MD5 verification

From 489ed0fc80bc3fde3daf0250042712cf000dfc0c Mon Sep 17 00:00:00 2001
From: David Jurado <david.jurado@factored.ai>
Date: Fri, 15 Nov 2024 10:18:03 -0500
Subject: [PATCH 14/14] Fix dependencies

---
 object_detection/Dockerfile.mlcube       | 14 ++---
 object_detection/download_demo.sh        |  4 +-
 object_detection/mlcube_requirements.txt | 77 ++++++++++++++++++++++++
 3 files changed, 83 insertions(+), 12 deletions(-)
 create mode 100644 object_detection/mlcube_requirements.txt

diff --git a/object_detection/Dockerfile.mlcube b/object_detection/Dockerfile.mlcube
index 5b802e3f9..009f22751 100644
--- a/object_detection/Dockerfile.mlcube
+++ b/object_detection/Dockerfile.mlcube
@@ -1,4 +1,4 @@
-FROM pytorch/pytorch:1.6.0-cuda10.1-cudnn7-devel
+FROM pytorch/pytorch:1.7.0-cuda11.0-cudnn8-devel
 
 RUN echo 'debconf debconf/frontend select Noninteractive' | debconf-set-selections
 RUN apt-key adv --fetch-keys https://developer.download.nvidia.com/compute/cuda/repos/ubuntu1804/x86_64/3bf863cc.pub
@@ -6,20 +6,14 @@ RUN apt-key adv --fetch-keys https://developer.download.nvidia.com/compute/cuda/
 # install basics
 RUN apt-get update -y \
  && apt-get install -y apt-utils curl unzip pv \
+                        libgl1-mesa-glx \
                        libglib2.0-0=2.56.1-2ubuntu1 \
                        libsm6=2:1.2.2-1 \
                        libxext6=2:1.3.3-1 \
                        libxrender-dev=1:0.9.10-1
 
-RUN pip install ninja==1.8.2.post2 \
-                yacs==0.1.5 \
-                cython==0.29.5 \
-                matplotlib==3.0.2 \
-                opencv-python==4.0.0.21 \
-                mlperf_compliance==0.0.10 \
-                torchvision==0.2.2 \
-                pycocotools==2.0.2 \
-                typer==0.3.2
+COPY mlcube_requirements.txt /root/mlcube_requirements.txt
+RUN pip install -r /root/mlcube_requirements.txt
 
 RUN pip install --no-cache-dir https://github.com/mlperf/logging/archive/9ea0afa.zip
 
diff --git a/object_detection/download_demo.sh b/object_detection/download_demo.sh
index 4a0fc10ac..86cd1964b 100755
--- a/object_detection/download_demo.sh
+++ b/object_detection/download_demo.sh
@@ -19,7 +19,6 @@ echo "Downloading annotations_trainval2017.zip:"
 curl -O https://mlcube.mlcommons-storage.org/minibenchmarks/object_detection.zip
 echo "Extracting demo_data.zip:"
 unzip -o -q object_detection.zip
-rm object_detection.zip
 echo "Done!"
 
 # MD5 verification
@@ -33,6 +32,7 @@ checkMD5() {
 }
 
 echo "validating demo_data.zip:"
-checkMD5 "demo_data.zip" "1b50202a21b0d8c3235d0a6f39b6f40c"
+checkMD5 "object_detection.zip" "1b50202a21b0d8c3235d0a6f39b6f40c"
+rm object_detection.zip
 
 popd
diff --git a/object_detection/mlcube_requirements.txt b/object_detection/mlcube_requirements.txt
new file mode 100644
index 000000000..1fe413916
--- /dev/null
+++ b/object_detection/mlcube_requirements.txt
@@ -0,0 +1,77 @@
+backcall==0.2.0
+beautifulsoup4==4.9.3
+certifi==2020.6.20
+cffi==1.14.0
+chardet==3.0.4
+click==8.1.7
+conda==4.9.1
+conda-build==3.20.5
+conda-package-handling==1.7.0
+contourpy==1.1.1
+cryptography==2.9.2
+cycler==0.12.1
+Cython==3.0.11
+dataclasses==0.6
+decorator==4.4.2
+dnspython==2.0.0
+filelock==3.0.12
+fonttools==4.54.1
+future==0.18.2
+glob2==0.7
+idna==2.9
+importlib-resources==6.4.5
+ipython==7.18.1
+ipython-genutils==0.2.0
+jedi==0.17.2
+Jinja2==2.11.2
+kiwisolver==1.4.7
+libarchive-c==2.9
+markdown-it-py==3.0.0
+MarkupSafe==1.1.1
+matplotlib==3.7.5
+mdurl==0.1.2
+mkl-fft==1.2.0
+mkl-random==1.1.1
+mkl-service==2.3.0
+mlperf-compliance==0.0.10
+ninja==1.11.1.1
+numpy==1.19.5
+olefile==0.46
+opencv-python==4.10.0.84
+packaging==24.2
+parso==0.7.0
+pexpect==4.8.0
+pickleshare==0.7.5
+Pillow==8.0.0
+pkginfo==1.6.0
+prompt-toolkit==3.0.8
+psutil==5.7.2
+ptyprocess==0.6.0
+pycocotools==2.0.7
+pycosat==0.6.3
+pycparser==2.20
+pygments==2.18.0
+pyOpenSSL==19.1.0
+pyparsing==3.1.4
+PySocks==1.7.1
+python-dateutil==2.9.0.post0
+python-etcd==0.4.5
+pytz==2020.1
+PyYAML==5.3.1
+requests==2.23.0
+rich==13.9.4
+ruamel-yaml==0.15.87
+shellingham==1.5.4
+six==1.14.0
+soupsieve==2.0.1
+torch==1.7.0
+torchelastic==0.2.1
+torchvision==0.8.0
+tqdm==4.46.0
+traitlets==5.0.5
+typer==0.13.0
+typing-extensions==3.7.4.3
+urllib3==1.25.8
+wcwidth==0.2.5
+yacs==0.1.8
+zipp==3.20.2