This project has two subtasks: category classification and pairwise compatibility classification. The first task requires construcing <img, class_label> pair for classification. The second task requires constructing <(x1,x2), binary_label> pair for classification.
- data preparation is defined in data.py
- CNN model in defined in model.py
- training is in train_*.py
- utils.py provides hyper-parameters
- pytorch/tensorflow>=2.0
- tqdm
- sklearn
- torchvision
- sklearn
Dataset:
- Polyvore Outfits (download link): https://drive.google.com/open?id=1ZCDRRh4wrYDq0O3FOptSlBpQFoe0zHTw
or directly download from google drive on AWS: https://gist.github.com/davidsonic/5c419db600f24a984d3a6b56b0b2053a