CDAL in PyTorch

PyTorch implementation of CDAL "Contextual Diversity for Active Learning" accepted in ECCV20.

Sharat Agarwal, Himanshu Arora, Saket Anand, Chetan Arora

First two authors contributed equally Link to the paper

Citation

If using this code, parts of it, or developments from it, please cite our paper:

@inproceedings{agarwal2020contextual,
  title={Contextual Diversity for Active Learning},
  author={Agarwal, Sharat and Arora, Himanshu and Anand, Saket and Arora, Chetan},
  booktitle={European Conference on Computer Vision},
  pages={137--153},
  year={2020},
  organization={Springer}
}

Proposed Architecture

Prerequisites:

• Python 3.6
• Pytorch >= 0.4.1
• CUDA 9.0 or higher
• CPU compatible but NVIDIA GPU + CUDA CuDNN is recommended.

Installation

Clone the repo:

$ git clone https://github.com/sharat29ag/CDAL
$ cd CDAL

Frame Selection

By default, logs are stored in <root_dir>/log with this structure:

<root_dir>/experiments/logs

Sample features in features folder for PASCAL-VOC.

For weighted features:

python preprocess.py

Change the path to raw features in the preprocess.py

Creates a folder <root_dir>/features2 with weighted features.

For CDAL-RL selection:

python main.py --number_of_picks <number of frames to select> --path_to_features <path to weighted features> --classes <number of classes in dataset> --gpu 1 --save-dir log/summe-split0 --start_idx 0

List of selected samples will be stored in <root_dir>/selection/

Base Networks

• Semantic Segmentation
• Object Detection
• Image Classification

Acknowledgements

This codebase is borrwed from VSUMM

Contact

If there are any questions or concerns feel free to send a message at [email protected]