Gender Artifacts in Visual Datasets

Project Page | Paper

This repo provides the code for the paper "Gender Artifacts in Visual Datasets."

  @article{meister2022artifacts,
  author = {Nicole Meister and Dora Zhao and Angelina Wang and Vikram V. Ramaswamy and Ruth Fong and Olga Russakovsky},
  title = {Gender Artifacts in Visual Datasetsi},
  journal = {CoRR},
  volume = {abs/2206.09191},
  year={2022}
  }

Setup

Setup computing environment

conda create -n genderartifacts python=3.9
conda activate genderartifacts 
conda install --file requirements.txt

Download data annotations

Download the annotations from the following sources and place them in data/{dataset_name}.

COCO

wget http://images.cocodataset.org/annotations/annotations_trainval2014.zip

OpenImages

Follow the instructions from the OpenImage website (copied below):

1. Download the downloader (open and press Ctrl + S), or directly run:

wget https://raw.githubusercontent.com/openimages/dataset/master/downloader.py

2. Run the following script, where $IMAGE_LIST_FILE is one of the files with image key lists above:

python downloader.py $IMAGE_LIST_FILE --download_folder=$DOWNLOAD_FOLDER --num_processes=5

Experiments

Resolution and Color

TBD

Person and Background

The files are located in src/person_scene. To generate the image manipulations in the paper, use the following scripts:

(* denotes available only for COCO)

Name	Script
Full NoBg	python image_manipulations.py --type full
MaskSegm*	python image_manipulations.py --type segm --background
MaskRect	python image_manipulations.py --type rect --background
MaskSegm NoBg*	python image_manipulations.py --type segm
MaskRect NoBg	python image_manipulations.py --type rect

Note: make sure to specify the arguments --dataset $DATA --filepath $PATH --annotations $ANN --split $SPLIT as well.

To train and evaluate the gender cue model, run the following scripts

Train: bash train.sh $TRAIN_LABEL_PATH $VAL_LABEL_PATH

Evaluate: bash eval.sh $MODEL_PATH $TEST_LABEL_PATH

Contextual Objects