SEAM Match-RCNN

Official code of MovingFashion: a Benchmark for the Video-to-Shop Challenge paper

Installation

Requirements:

• Pytorch 1.5.1 or more recent, with cudatoolkit (10.2)
• torchvision
• tensorboard
• cocoapi
• OpenCV Python
• tqdm
• cython
• CUDA >= 10

Step-by-step installation

# first, make sure that your conda is setup properly with the right environment
# for that, check that `which conda`, `which pip` and `which python` points to the
# right path. From a clean conda env, this is what you need to do

conda create --name seam -y python=3
conda activate seam

pip install cython tqdm opencv-python

# follow PyTorch installation in https://pytorch.org/get-started/locally/
conda install pytorch torchvision cudatoolkit=10.2 -c pytorch

conda install tensorboard

export INSTALL_DIR=$PWD

# install pycocotools
cd $INSTALL_DIR
git clone https://github.com/cocodataset/cocoapi.git
cd cocoapi/PythonAPI
python setup.py build_ext install

# download SEAM
cd $INSTALL_DIR
git clone https://github.com/VIPS4/SEAM-Match-RCNN.git
cd SEAM-Match-RCNN
mkdir data
mkdir ckpt

unset INSTALL_DIR

Dataset

SEAM Match-RCNN has been trained and test on MovingFashion and DeepFashion2 datasets. Follow the instruction to download and extract the datasets.

We suggest to download the datasets inside the folder data.

MovingFashion

MovingFashion dataset is available for academic purposes here.

Deepfashion2

DeepFashion2 dataset is available here. You need fill in the form to get password for unzipping files.

Once the dataset will be extracted, use the reserved DeepFtoCoco.py script to convert the annotations in COCO format, specifying dataset path.

python DeepFtoCoco.py --path <dataset_root>

Training

We provide the scripts to train both Match-RCNN and SEAM Match-RCNN. Check the scripts for all the possible parameters.

Single GPU

#training of Match-RCNN
python train_matchrcnn.py --root_train <path_of_images_folder> --train_annots <json_path> --save_path <save_path> 

#training on movingfashion
python train_movingfashion.py --root <path_of_dataset_root> --train_annots <json_path> --test_annots <json_path> --pretrained_path <path_of_matchrcnn_model>


#training on multi-deepfashion2
python train_multiDF2.py --root <path_of_dataset_root> --train_annots <json_path> --test_annots <json_path> --pretrained_path <path_of_matchrcnn_model>

Multi GPU

We use internally torch.distributed.launch in order to launch multi-gpu training. This utility function from PyTorch spawns as many Python processes as the number of GPUs we want to use, and each Python process will only use a single GPU.

#training of Match-RCNN
python -m torch.distributed.launch --nproc_per_node=<NUM_GPUS> train_matchrcnn.py --root_train <path_of_images_folder> --train_annots <json_path> --save_path <save_path>

#training on movingfashion
python -m torch.distributed.launch --nproc_per_node=<NUM_GPUS> train_movingfashion.py --root <path_of_dataset_root> --train_annots <json_path> --test_annots <json_path> --pretrained_path <path_of_matchrcnn_model> 

#training on multi-deepfashion2
python -m torch.distributed.launch --nproc_per_node=<NUM_GPUS> train_multiDF2.py --root <path_of_dataset_root> --train_annots <json_path> --test_annots <json_path> --pretrained_path <path_of_matchrcnn_model> 

Pre-Trained models

It is possibile to start training using the MatchRCNN pre-trained model.

[MatchRCNN] Pre-trained model on Deepfashion2 is available to download here. This model can be used to start the training at the second phase (training directly SEAM Match-RCNN).

We suggest to download the model inside the folder ckpt.

Evaluation

To evaluate the models of SEAM Match-RCNN please use the following scripts.

#evaluation on movingfashion
python evaluate_movingfashion.py --root_test <path_of_dataset_root> --test_annots <json_path> --ckpt_path <checkpoint_path>


#evaluation on multi-deepfashion2
python evaluate_multiDF2.py --root_test <path_of_dataset_root> --test_annots <json_path> --ckpt_path <checkpoint_path>

Citation

@misc{godi2021movingfashion,
      title={MovingFashion: a Benchmark for the Video-to-Shop Challenge}, 
      author={Marco Godi and Christian Joppi and Geri Skenderi and Marco Cristani},
      year={2021},
      eprint={2110.02627},
      archivePrefix={arXiv},
      primaryClass={cs.CV}
}

This work is licensed under a Creative Commons Attribution-NonCommercial-ShareAlike 4.0 International License.