mica-MovieCLIP

This repository contains the codebase for MovieCLIP: Visual Scene Recognition in Movies

Installation

• Install the environment for training the baseline LSTM models using the following commands:

  conda create -n py37env python=3.7
  conda install pytorch==1.7.1 torchvision==0.8.2 torchaudio==0.7.2 cudatoolkit=10.1 -c pytorch
  pip install -r requirements.txt --use-deprecated=legacy-resolver
  

• Install CLIP dependencies using the following commands:

  pip install ftfy regex tqdm
  pip install git+https://github.com/openai/CLIP.git
  

Data setup

• Please refer to README.md under the data_splits folder for instructions on using the MovieCLIP dataset.

Visual scene tagging

• Please refer to README.md under the preprocess_scripts/visual_scene_tagging folder for instructions on using the CLIP model for tagging the visual scenes in the MovieCLIP dataset.

To Dos

• Add the dataset link and instructions for using the MovieCLIP dataset
• Add code for tagging using the CLIP model
• Add code for training the baseline LSTM models
• Add code for openmmlab setup and Swin-B model inference

If you find this repository useful, please cite the following paper:

@InProceedings{Bose_2023_WACV,
    author    = {Bose, Digbalay and Hebbar, Rajat and Somandepalli, Krishna and Zhang, Haoyang and Cui, Yin and Cole-McLaughlin, Kree and Wang, Huisheng and Narayanan, Shrikanth},
    title     = {MovieCLIP: Visual Scene Recognition in Movies},
    booktitle = {Proceedings of the IEEE/CVF Winter Conference on Applications of Computer Vision (WACV)},
    month     = {January},
    year      = {2023},
    pages     = {2083-2092}
}

For any questions, please open an issue and feel free to contact Digbalay Bose (dbose@usc.edu)