Project Fashion mmfashion

Get started Recommender

Environment

• conda
  • conda

      conda create -n fashion python=3.8 
    

  • install pytorch

      conda install pytorch torchvision torchaudio cudatoolkit=10.2 -c pytorch
    

  • install mmcv

      pip install mmcv-full -f https://download.openmmlab.com/mmcv/dist/cu100/torch1.9.0/index.html
    

Demo code

Error

attribute predict

• FileNotFoundError: file “/DL/ProjFashion/mmfashion-master/configs/attribute_predict/global_predictor_vgg_attr.py” does not exist
  • test_attr_predictor.py
    • line39:
      • from: default='configs/attribute_predict/global_predictor_vgg_attr.py'
      • to: default='configs/attribute_predict_coarse/global_predictor_vgg_attr.py'

fashion compatibility and recommendation

• OSError: checkpoint/resnet18.pth is not a checkpoint file

    import torch
    import torchvision
  
    model = torchvision.models.resnet18(pretrained=True, progress=True)
    torch.save(model.state_dict(), 'checkpoint/resnet18.pth')
  

• OSError: checkpoint/FashionRecommend/TypeAware/disjoint/l2_embed/epoch_16.pth is not a checkpoint file
  • mkdir -p checkpoint/FashionRecommend/TypeAware/disjoint/l2_embed/
  • and download weight from here
    • This one below 👇
    • ResNet-18 Disjoint fully-connected layer Triplet loss, Type-specific loss, Similarity loss, VSE loss 50.4 0.80
• NotImplementedError: There were no tensor arguments to this function (e.g., you passed an empty list of Tensors), but no fallback function is registered for schema aten::_cat. This usually means that this function requires a non-empty list of Tensors, or that you (the operator writer) forgot to register a fallback function. Available functions are [CPU, CUDA, QuantizedCPU, BackendSelect, Named, ADInplaceOrView, AutogradOther, AutogradCPU, AutogradCUDA, AutogradXLA, UNKNOWN_TENSOR_TYPE_ID, AutogradMLC, AutogradHPU, AutogradNestedTensor, AutogradPrivateUse1, AutogradPrivateUse2, AutogradPrivateUse3, Tracer, Autocast, Batched, VmapMode].
  • this means no data
  • they provide this code but they deleted set2, so use set1 or set3

      python demo/test_fashion_recommender.py \
      --input_dir demo/imgs/fashion_compatibility/set2
    

      python demo/test_fashion_recommender.py \
      --input_dir demo/imgs/fashion_compatibility/set3
    

• FileNotFoundError: [Errno 2] No such file or directory: 'data/Polyvore/polyvore_item_metadata.json'
  • download polyvore data
  • mkdir -p data/Polyvore/
  • for predict, only json file here
• FileNotFoundError: [Errno 2] No such file or directory: 'data/Polyvore/disjoint/test.json'
  • download polyvore data
  • mkdir -p data/Polyvore/disjoint
• FileNotFoundError: [Errno 2] No such file or directory: 'data/Polyvore/disjoint/typespaces.p'
  • duplicated
• FileNotFoundError: [Errno 2] No such file or directory: 'data/Polyvore/disjoint/compatibility_test.txt'
  • duplicated
• Done.

Compatibility Score

only shows:
Compatibility score: 0.292

Issue: You can offer a set of possible choices. And pick up the highest scored one. https://github.com/open-mmlab/mmfashion/issues/64

BERT4REC

https://aihub.or.kr/sites/default/files/Sample_data/%EA%B5%AC%EC%B6%95%ED%99%9C%EC%9A%A9%EA%B0%80%EC%9D%B4%EB%93%9C%EB%B6%81_2020-01/014.K_Fashion_%EC%9D%B4%EB%AF%B8%EC%A7%80_%EB%8D%B0%EC%9D%B4%ED%84%B0_%EA%B5%AC%EC%B6%95_%EA%B0%80%EC%9D%B4%EB%93%9C%EB%9D%BC%EC%9D%B8.pdf
http://dsba.korea.ac.kr/seminar/?mod=document&uid=48
https://arxiv.org/abs/1904.06690

Get started Segmentation

README.md

Appendix

Reference

https://github.com/open-mmlab/mmfashion
https://github.com/open-mmlab/mmcv