Official PyTorch reimplementation of "Adaptation of Manga Face Representation for Accurate Clustering" presented in SIGGRAPH Asia 2018.
pip install -r requirements.txtWe conduct experiments using torch==1.8.0+cu111 and torchvision==0.9.0+cu111.
Download the Manga109 dataset from the official website.
The annotation version that we used is v2018.05.31.
You can see the version list here.
# script/split_dataset.py /path/to/Manga109_20xx_xx_xx
python script/crop_faces.py /path/to/Manga109_20xx_xx_xx
# script/split_dataset.py /path/to/Manga109_20xx_xx_xxThe commands that are commented out have already been executed.
The statistics of the test data can be obtained using the below command.
python script/get_data_stats.py /path/to/Manga109_20xx_xx_xxIf you want to pre-train a model by yourself, execute the following command.
python pre_train.py /path/to/Manga109_20xx_xx_xx
# evaluation of a pre-trained model
python eval.py /path/to/Manga109_20xx_xx_xx --model_path results/model.pthIf not, download our pre-trained model from here and put it to results/model.pth.
Set a value in [0, 10] as an argument of --title_idx.
python train.py /path/to/Manga109_20xx_xx_xx --title_idx 0| Method | Accuracy | NMI |
|---|---|---|
| Pre-train (paper) | 0.48 | 0.63 |
| Pre-train (reimpl.) | 0.472 | 0.615 |
| Fine-tune (paper) | 0.64 | 0.71 |
| Fine-tune (reimpl.) | 0.666 | 0.718 |
- We implemented with PyTorch instead of Chainer. This is because ResNet-50 pre-trained on ImageNet for Chainer is not publicly available.
- The size of the dataset for pre-training (67,336) is somewhat larger than the value in our paper (67,328).
- The size of BEMADER_P (1,111 + 82) is somewhat larger than the value in our paper (1,105 + 82).
- Our paper: https://dl.acm.org/citation.cfm?id=3283319
- manga109api: https://github.com/manga109/manga109api