Home

Awesome

Video Corpus Moment Retrieval with Contrastive Learning

PyTorch implementation for the paper "Video Corpus Moment Retrieval with Contrastive Learning" (SIGIR 2021, long paper): SIGIR version, ArXiv version.

model_overview

The codes are modified from TVRetrieval.

Prerequisites

If you have Anaconda installed, the conda environment of ReLoCLNet can be built as follows (take python 3.7 as an example):

conda create --name reloclnet python=3.7
conda activate reloclnet
conda install -c anaconda cudatoolkit cudnn  # ignore this if you already have cuda installed
conda install pytorch==1.7.0 torchvision==0.8.0 torchaudio==0.7.0 cudatoolkit=11.0 -c pytorch
conda install -c anaconda h5py=2.9.0
conda install -c conda-forge transformers tensorboard tqdm easydict

The conda environment of TVRetrieval also works.

Getting started

  1. Clone this repository
$ git clone git@github.com:IsaacChanghau/ReLoCLNet.git
$ cd ReLoCLNet
  1. Download features

For the features of TVR dataset, please download tvr_feature_release.tar.gz (link is copied from TVRetrieval#prerequisites) and extract it to the data directory:

$ tar -xf path/to/tvr_feature_release.tar.gz -C data

This link may be useful for you to directly download Google Drive files using wget. Please refer TVRetrieval#prerequisites for more details about how the features are extracted if you are interested.

  1. Add project root to PYTHONPATH (Note that you need to do this each time you start a new session.)
$ source setup.sh

Training and Inference

TVR dataset

# train, refer `method_tvr/scripts/train.sh` and `method_tvr/config.py` more details about hyper-parameters
$ bash method_tvr/scripts/train.sh tvr video_sub_tef resnet_i3d --exp_id reloclnet
# inference
# the model directory placed in method_tvr/results/tvr-video_sub_tef-reloclnet-*
# change the MODEL_DIR_NAME as tvr-video_sub_tef-reloclnet-*
# SPLIT_NAME: [val | test]
$ bash method_tvr/scripts/inference.sh MODEL_DIR_NAME SPLIT_NAME

For more details about evaluation and submission, please refer TVRetrieval#training-and-inference.

Citation

If you feel this project helpful to your research, please cite our work.

@inproceedings{zhang2021video,
	author = {Zhang, Hao and Sun, Aixin and Jing, Wei and Nan, Guoshun and Zhen, Liangli and Zhou, Joey Tianyi and Goh, Rick Siow Mong},
	title = {Video Corpus Moment Retrieval with Contrastive Learning},
	year = {2021},
	isbn = {9781450380379},
	publisher = {Association for Computing Machinery},
	address = {New York, NY, USA},
	url = {https://doi.org/10.1145/3404835.3462874},
	doi = {10.1145/3404835.3462874},
	booktitle = {Proceedings of the 44th International ACM SIGIR Conference on Research and Development in Information Retrieval},
	pages = {685–695},
	numpages = {11},
	location = {Virtual Event, Canada},
	series = {SIGIR '21}
}

TODO