Repository files navigation

We introduce a large-scale image datasetEasyPortrait for portrait segmentation and face parsing. Proposed dataset can be used in several tasks, such as background removal in conference applications, teeth whitening, face skin enhancement, red eye removal or eye colorization, and so on.

EasyPortrait dataset size is about91.78GB, and it contains40,000 RGB images (~38.3K FullHD images) with high quality annotated masks. This dataset is divided into training set, validation set and test set by subjectuser_id. The training set includes 30,000 images, the validation set includes 4,000 images, and the test set includes 6,000 images.

For more information see our paperEasyPortrait – Face Parsing and Portrait Segmentation Dataset.

• 2023/11/13: We release EasyPortrait 2.0. ✌️
  • 40,000 RGB images (~38.3K FullHD images)
  • Added diversity by region, race, human emotions and lighting conditions
  • The data was further cleared and new ones were added
  • Train/val/test split: (30,000)75% / (4,000)10% / (6,000)15% by subjectuser_id
  • Multi-gpu training and testing
  • Added new models for face parsing and portrait segmentation
  • Dataset size is91.78GB
  • 13,705 unique persons
• 2023/02/23: EasyPortrait (Initial Dataset) 💪
  • Dataset size is26GB
  • 20,000 RGB images (~17.5K FullHD images) with9 classes annotated
  • Train/val/test split: (14,000)70% / (2,000)10% / (4,000)20% by subjectuser_id
  • 8,377 unique persons

Old EasyPortrait dataset is also available into branchEasyPortrait_v1!

Also, you can download EasyPortrait dataset fromKaggle.

.├── images.zip│   ├── train/         # Train set: 30k│   ├── val/           # Validation set: 4k│   ├── test/          # Test set: 6k├── annotations.zip│   ├── train/│   ├── val/│   ├── test/├── meta.zip       # Meta-information (width, height, brightness, imhash, user_id)...

We provide some pre-trained models as the baseline for portrait segmentation and face parsing. We use mean Intersection over Union (mIoU) as the main metric.

Annotations are presented as 2D-arrays, images in*.png format with several classes:

Also, we provide some additional meta-information for dataset inannotations/meta.zip file:

where:

• image_name - image file name without extension
• user_id - unique anonymized user ID
• height - image height
• width - image width
• brightness - image brightness
• set - "train", "test" or "val" for train / test / val subsets respectively

The code is based onMMSegmentation with 0.30.0 version.

Models were trained and evaluated on 8 NVIDIA V100 GPUs with CUDA 11.2.

For installation process follow the instructionshere and use therequirements.txt file in our repository.

python ./pipelines/tools/train.py ./pipelines/local_configs/easy_portrait_experiments/<model_dir>/<config_file>.py --gpu-id <GPU_ID>

./pipelines/tools/dist_train.sh ./pipelines/local_configs/easy_portrait_experiments/<model_dir>/<config_file>.py <NUM_GPUS>

python ./pipelines/tools/test.py <PATH_TO_MODEL_CONFIG>  <PATH_TO_CHECKPOINT> --gpu-id <GPU_ID> --eval mIoU

./pipelines/tools/dist_test.sh <PATH_TO_MODEL_CONFIG>  <PATH_TO_CHECKPOINT> <NUM_GPUS> --eval mIoU

python ./pipelines/demo/image_demo.py <PATH_TO_IMG> <PATH_TO_MODEL_CONFIG> <PATH_TO_CHECKPOINT> --palette=easy_portrait --out-file=<PATH_TO_OUT_FILE>

• Alexander Kapitanov
• Karina Kvanchiani
• Elizaveta Petrova
• Karen Efremyan
• Alexander Sautin

• arXiv
• Paperswithcode
• Kaggle
• Habr
• Gitlab

You can cite the paper using the following BibTeX entry:

@article{EasyPortrait,    title={EasyPortrait - Face Parsing and Portrait Segmentation Dataset},    author={Kapitanov, Alexander and Kvanchiani, Karina and Kirillova Sofia},    journal={arXiv preprint arXiv:2304.13509},    year={2023}}

This work is licensed under a variant ofCreative Commons Attribution-ShareAlike 4.0 International License.

Please see the specificlicense.