Setup

Expected data structure for DeepFashion2

Root dataset is default to data/

data/
    deepfashion2/
        train/
        val/
        train_triplets.csv
        val_triplets.csv

Create environment

In root (repo) directory, create and activate environment using pip:

python -m venv ./vs
source ./vs/bin/activate
pip install -r requirements.txt
pip install -e .

Train Scripts

Scripts are located under scripts/ Edit your arguments and hyperparameters in each .sh file.

Before training, login with wandb to monitor logs. Enter your API key at Authorize Page

wandb login
<COPY AND ENTER API KEY>

Train with script file

Make file executable and run in shell

chmod +x ./scripts/train.sh
./train.sh

Train directly

python src/train.py \
--project-name visual-search --exp-name train_v1_lr1e-3_bsz1_e20 \
--backbone resnet50