Code and dataset for the ICCV 2023 paper:
Data-Free Class-Incremental Hand Gesture Recognition
S. Aich, J Ruiz-Santaquiteria, Z. Lu, P. Garg, K J Joseph, A. F. Garcia, V. N. Balasubramanian, K. Kin, C. Wan, N. C. Camgoz, S. Ma, and F De la Torre
International Conference on Computer Vision (ICCV), 2023
[project]
This (main) branch contains the implementation of the method proposed in this paper. Please refer to the baseline branch for the other DFCIL methods used for benchmarking.
-
Make sure the directories are correct for you in the ./scripts/common_dirs.sh file.
-
Go to the corresponding scripts directory for a particular dataset:
$ cd scripts/<DATASET>/mi_drop/supcon
$ # <DATASET> is hgr_shrec_2017 or ego_gesture
- Run with a particular seed:
$ ./all_runner_seed.sh <SEED> <GPU_ID>
$ ./all_runner_seed.sh 0 1 # seed 0 gpu 1
-
The accuracies with incremental class indices will be logged into the ./scripts//mi_drop/supcon/tmp directory.
-
Good luck!
- EgoGesture3D: The license for the EgoGesture3D skeletons are the same as this repository (license.txt). Please refer to the EgoGesture paper and the website for the original video dataset and corresponding license.
- SHREC-2017 train/val/test splits: This zip file only contains the split files comprising the list of files. Please refer to the SHREC 2017 website to download the dataset.
If you find this paper/code useful, please consider citing (.bib):
@InProceedings{boat-mi-dfcil,
author = {Aich, Shubhra and Ruiz-Santaquiteria, Jesus and Lu, Zhenyu and Garg, Prachi and Joseph, KJ and Fernandez Garcia, Alvaro and Balasubramanian, Vineeth N and Kin, Kenrick and Wan, Chengde and Camgoz, Necati Cihan and Ma, Shugao and De la Torre, Fernando},
title = {Data-Free Class-Incremental Hand Gesture Recognition},
booktitle = ICCV,
year = {2023},
}