CODAE - Co-occurence Denoising Autoencoder

    git clone https://github.com/victordeleau/codae
    cd codae/
    conda create -f env.yml
    conda activate codae
    pip install -e ./

Steps

4 steps are involved. They can be started independently, or all at once using

python3 script/4_complementarity_inference

I - Segment dataset

python3 script/1_segment_dataset.py

II - Encode dataset

python3 script/2_encode_dataset.py

III - Train autoencoder

python3 script/3_train_autoencoder.py

IV - Complementarity inference

python3 script/4_complementarity_inference.py

Dataset requirement

• Full body
• At least 3 categories (top/bottom/shoes)
• Simple clothes
• Men only

Dataset characteristics

DeepFashion2

Number of valid segmentation mask: Number of images with more than 3 valid segmentation mask: 3324