/Perfect-Half-Million-Beauty-Product-Image-Recognition-Challenge-2019

the 2nd place solution (2019)

Primary LanguagePythonApache License 2.0Apache-2.0

Perfect Half Million Beauty Product Image Recognition Challenge

This is my solution for Perfect Half Million Beauty Product Image Recognition Challenge , which obtained the 2nd place (007) with MAP@7 0.407233.

Requirement

  • python 3.6
  • pytorch 0.4.0
  • PIL
  • torchvision here
  • pretrainedmodels
  • faiss here
  • download image data from here if published and place them in ./data/ and ./test/
  • place val.csv and test.csv in ./

How to use

  • For obtaining the retrieval result, run bash ./predict
    which needs three inputs: "train_images_path", "test_images_path", "predictions_path"
  • For training UEL, run python trainUEL.py
    which needs three inputs: "train_path", "test_path", "tlabel"
    • -train_path (default: 1) : the file for training data
    • -test_path (default: './data/') : the file for val data
    • -tlabel (default: './val.csv') : the label for val data

Please cite the paper if you are using this code.

Zehang Lin, Haoran Xie, Peipei Kang, Zhenguo Yang, Wenyin Liu and Qing Li. Cross-domain Beauty Item Retrieval via Unsupervised Embedding Learning. Proceedings of the 27th ACM International Conference on Multimedia, 2019: 2543-2547.