This is part implementation of Harmonizing Base and Novel Classes: A Class-Contrastive Approach for Generalized Few-Shot Segmentation (IJCV 2023).
The training configurations for Pascal-VOC and MS COCO datasets adhere to the standards established in the original CAPL methodology(https://github.com/dvlab-research/GFS-Seg).
This code reads data from .txt files where each line contains the paths for image and the correcponding label respectively. Image and label paths are seperated by a space. Example is as follows:
image_path_1 label_path_1
image_path_2 label_path_2
image_path_3 label_path_3
...
image_path_n label_path_n
Then update the train/val list paths in the config files.
-
For training, please set the option only_evaluate to False in the configuration file. Then execute this command at the root directory:
sh train.sh {dataset} {model_config}
-
For evaluation only, please set the option only_evaluate to True in the corresponding configuration file.
Example: Train / Evaluate CAPL with 1-shot on the split 0 of PASCAL-5i:
sh train.sh pascal split0_1shot
Our work is closely related to the following assets that inspire our implementation. We gratefully thank the authors.
If you find this project useful, please consider citing:
@article{liu2023harmonizing,
title={Harmonizing Base and Novel Classes: A Class-Contrastive Approach for Generalized Few-Shot Segmentation},
author={Liu, Weide and Wu, Zhonghua and Zhao, Yang and Fang, Yuming and Foo, Chuan-Sheng and Cheng, Jun and Lin, Guosheng},
journal={International Journal of Computer Vision},
year={2023},
publisher={Springer}
}
- Full implementation of the code in the paper.
- Code refactoring