release

README.md

@@ -0,0 +1,82 @@
+# GraspNet graspness
+My implementation of paper "Graspness Discovery in Clutters for Fast and Accurate Grasp Detection" (ICCV 2021).
+
+[[paper](https://openaccess.thecvf.com/content/ICCV2021/papers/Wang_Graspness_Discovery_in_Clutters_for_Fast_and_Accurate_Grasp_Detection_ICCV_2021_paper.pdf)]
+[[dataset](https://graspnet.net/)]
+[[API](https://github.com/graspnet/graspnetAPI)]
+
+
+## Requirements
+- Python 3
+- PyTorch 1.6
+- Open3d 0.8
+- TensorBoard 2.3
+- NumPy
+- SciPy
+- Pillow
+- tqdm
+- MinkowskiEngine
+
+## Installation
+Get the code.
+```bash
+git clone https://github.com/rhett-chen/graspness_implementation.git
+cd graspnet-graspness
+```
+Install packages via Pip.
+```bash
+pip install -r requirements.txt
+```
+Compile and install pointnet2 operators (code adapted from [votenet](https://github.com/facebookresearch/votenet)).
+```bash
+cd pointnet2
+python setup.py install
+```
+Compile and install knn operator (code adapted from [pytorch_knn_cuda](https://github.com/chrischoy/pytorch_knn_cuda)).
+```bash
+cd knn
+python setup.py install
+```
+Install graspnetAPI for evaluation.
+```bash
+git clone https://github.com/graspnet/graspnetAPI.git
+cd graspnetAPI
+pip install .
+```
+
+## Point level Graspness Generation
+Point level graspness label are not included in the original dataset, and need additional generation. Make sure you have downloaded the orginal dataset from [GraspNet](https://graspnet.net/). The generation code is in [dataset/generate_graspness.py](dataset/generate_graspness.py).
+```bash
+cd dataset
+python generate_graspness.py --dataset_root /data3/graspnet --camera_type kinect
+```
+
+## Simplify dataset
+original dataset grasp_label files have redundant data,  We can significantly save the memory cost. The code is in [dataset/simplify_dataset.py](dataset/simplify_dataset.py)
+```bash
+cd dataset
+python simplify_dataset.py --dataset_root /data3/graspnet
+```
+
+## Training and Testing
+Training examples are shown in [command_train.sh](command_train.sh). `--dataset_root`, `--camera` and `--log_dir` should be specified according to your settings. You can use TensorBoard to visualize training process.
+
+Testing examples are shown in [command_test.sh](command_test.sh), which contains inference and result evaluation. `--dataset_root`, `--camera`, `--checkpoint_path` and `--dump_dir` should be specified according to your settings. Set `--collision_thresh` to -1 for fast inference.
+
+If you need the trained weights, you can contact me directly.
+
+## Results
+Results "In repo" report the model performance of my results without collision detection.
+
+Evaluation results on Kinect camera:
+|          |        | Seen             |                  |        | Similar          |                  |        | Novel            |                  | 
+|:--------:|:------:|:----------------:|:----------------:|:------:|:----------------:|:----------------:|:------:|:----------------:|:----------------:|
+|          | __AP__ | AP<sub>0.8</sub> | AP<sub>0.4</sub> | __AP__ | AP<sub>0.8</sub> | AP<sub>0.4</sub> | __AP__ | AP<sub>0.8</sub> | AP<sub>0.4</sub> |
+| In paper | 61.19  | 71.46            | 56.04            | 47.39  | 56.78            | 40.43            | 19.01  | 23.73            | 10.60             |
+| In repo  | 61.83  | 73.28            | 54.14            | 51.13  | 62.53            | 41.57            | 19.94  | 24.90            | 11.02             |
+
+## Troubleshooting
+If you meet the torch.floor error in MinkowskiEngine, you can simplify solve it by change the source code of MinkowskiEngine: 
+MinkowskiEngine/utils/quantization.py 262，from discrete_coordinates =_auto_floor(coordinates) to discrete_coordinates = coordinates
+## Acknowledgement
+My code is mainly based on Graspnet-baseline  https://github.com/graspnet/graspnet-baseline.

command_test.sh

@@ -0,0 +1 @@
+CUDA_VISIBLE_DEVICES=4 python test.py --camera kinect --dump_dir logs/log_kn/dump_epoch03 --checkpoint_path logs/log_kn/epoch03.tar --batch_size 1 --dataset_root /data3/graspnet --infer --eval

command_train.sh

@@ -1 +1 @@
-CUDA_VISIBLE_DEVICES=4 python train.py --log_dir logs/log_kn_v9 --batch_size 4 --learning_rate 0.001 --model_name np15000_graspness1e-1_bs4_lr1e-3_viewres_dataaug_fps_14C --dataset_root /data3/graspnet
+CUDA_VISIBLE_DEVICES=4 python train.py --camera kinect --log_dir logs/log_kn --batch_size 4 --learning_rate 0.001 --model_name minkuresnet --dataset_root /data3/graspnet

dataset/generate_graspness.py

@@ -10,17 +10,24 @@
 import torch
 from graspnetAPI.utils.xmlhandler import xmlReader
 from graspnetAPI.utils.utils import get_obj_pose_list, transform_points
+import argparse
+
+
+parser = argparse.ArgumentParser()
+parser.add_argument('--dataset_root', default=None, required=True)
+parser.add_argument('--camera_type', default='kinect', help='Camera split [realsense/kinect]')
 
 
 if __name__ == '__main__':
-    dataset_root = '/media/bot/980A6F5E0A6F38801/datasets/graspnet/'
-    camera_type = 'realsense'
+    cfgs = parser.parse_args()
+    dataset_root = cfgs.dataset_root   # set dataset root
+    camera_type = cfgs.camera_type   # kinect / realsense
     save_path_root = os.path.join(dataset_root, 'graspness')
 
     num_views, num_angles, num_depths = 300, 12, 4
     fric_coef_thresh = 0.6
     point_grasp_num = num_views * num_angles * num_depths
-    for scene_id in range(101):
+    for scene_id in range(100):
         for ann_id in range(256):
             # get scene point cloud
             print('generating scene: {} ann: {}'.format(scene_id, ann_id))