Skip to content

dhw/Match-R-CNN-Repoduction

 
 

Repository files navigation

Match-R-CNN-Repoduction

Reproducing the fashion image categorization and retrieval baseline approach from https://github.com/switchablenorms/DeepFashion2 / https://arxiv.org/pdf/1901.07973.pdf

Installation

  • conda install --file requirements.txt

Procedure

Data-preparation

  • download deepfashion2 dataset
  • run src/data/split_dataset.py on train dataset
  • run src/data/deepfashion2_to_coco.py with train, validation and test dataset

Train feature-network

  • run src/models/train_feature_network.py with train dataset

Feature and pair creation

  • run src/features/get_features.py with train, validation and test dataset
  • run src/data/create_all_pairs.py with train, validation and test dataset
  • run src/make_item_level_pairs.py with train, validation and test dataset

Train matching-network

  • run src/models/train_model.py

About

Reproducing the fashion image categorization and retrieval baseline approach from https://github.com/switchablenorms/DeepFashion2 / https://arxiv.org/pdf/1901.07973.pdf

Resources

License

Stars

Watchers

Forks

Releases

No releases published

Packages

No packages published

Languages

  • Python 74.0%
  • Jupyter Notebook 16.7%
  • Makefile 9.3%