Implementation of paper - Outfit Transformer: Outfit Representations for Fashion Recommendation
โ ๏ธ The original paper outlines the specifics of the target item for Compatitible Item Retrieval (CIR) and Fill-in-the-Blank (FITB). Nonetheless, for the sake of impartial evaluation alongside other models, this information was intentionally excluded. (Should a dataset emerge that necessitates the prediction of a matching item when presented with a description unrelated to the target item itself, the model will be retrained accordingly.)
The figures below are derived using the Polyvore test dataset.
This code is tested with python 3.9.16, torch 1.12.1
python -m pip install -r requirements.txt
Download the polyvore dataset from here
python train.py --task cp --train_batch 64 --valid_batch 96 --n_epochs 5 --learning_rate 1e-5 --scheduler_step_size 1000 --work_dir $WORK_DIR --data_dir $DATA_DIR --wandb_api_key $WANDB_API_KEY
python train.py --task cir --train_batch 64 --valid_batch 96 --n_epochs 5 --learning_rate 1e-5 --scheduler_step_size 1000 --work_dir $WORK_DIR --data_dir $DATA_DIR --wandb_api_key $WANDB_API_KEY --checkpoint $CHECKPOINT
python test.py --task $TASK --polyvore_split nondisjoint --test_batch 96 --data_dir $DATA_DIR --checkpoint $CHECKPOINT
Download the checkpoint from here