Tool for getting the dataset of cropped faces from e621. It was created by training a YOLOv3 network on annotated facial features from about 1500 faces.
The total dataset includes ~186k faces. Rather than provide the cropped images, this repo contains CSV files with the bounding boxes of the detected features from my trained network, and a script to download the images from e621 and crop them based on these CSVs.
The CSVs also contain a subset of tags, which could potentially be used as labels to train a conditional GAN.
| File | |
|---|---|
| get_faces.py | Script for downloading base e621 files and cropping them based on the coordinates in the CSVs. |
| faces_s.csv | CSV containing URLs, bounding boxes, and a subset of the tags for 90k cropped faces with rating=safe from e621. |
| features_s.csv | CSV containing the bounding boxes for 389k facial features with rating=safe from e621. |
| faces_q.csv | CSV containing URLs, bounding boxes, and a subset of the tags for 96k cropped faces with rating=questionable from e621. |
| features_q.csv | CSV containing the bounding boxes for 400k facial features with rating=questionable from e621. |
The header for the faces_(s|q).csv files is:
e621id,feature,index,confidence,xmin,ymin,xmax,ymax,file_url,rating,score,file_size,tags,artist,copyrights,characters,species
And the header for the features_(s|q).csv files is:
e621id,feature,index,confidence,xmin,ymin,xmax,ymax
If you use Anaconda, you can do this to set up the env. The only requirements are wget and opencv2, so this probably isn't super necessary though. ¯\_(ツ)_/¯
conda env create -f e621_faces.yml
conda activate e621_faces
Before getting the full dataset, it's usually a good idea to test out the settings on a subset.
Run:
python get_faces.py download --min-score=250 --csv=faces_s.csv --species=canine
It should create a folder called out and download 14 images.
Then do:
python get_faces.py crop --min-score=250 --csv=faces_s.csv --species=canine --square
It should create a folder called crop with 17 cropped 512x512 images.
The dataset from faces_s.csv requires ~106GB to download the uncropped images. You can pull the full rating:s dataset as so:
python get_faces.py download --min-score=0 --min-confidence=0.0 --csv=faces_s.csv
python get_faces.py crop --min-score=0 --min-confidence=0.0 --csv=faces_s.csv
If you're planning to train a GAN using the images, you should probably only get images with confidence >= 0.99.
If you want to filter based on other attributes, you can modify the code to filter the faces DataFrame before it gets passed to download_images and crop_images.
The cropped images should all be formatted correctly so that you can run the StyleGAN dataset_tool.py on the directory without problems.
You can use the crop coordinates to train a network to detect faces using darknet. To do this, you would need to create a .txt file for each uncropped image, with the following info for each face on a new line:
<object-class> <x_center> <y_center> <width> <height>
and then run the training command as per the instructions in the above link.
