https://github.com/cskyl/SAM_WSSS

SAM Enhance Mask Quality for WSSS: This repository provides tools for generating, evaluating, and visualizing enhanced pseudo masks for Weakly Supervised Semantic Segmentation (WSSS) using the Segment Anything Model (SAM).
https://github.com/cskyl/SAM_WSSS

pascal-voc sam weakly-supervised-learning wsss

Last synced: about 2 months ago
JSON representation

SAM Enhance Mask Quality for WSSS: This repository provides tools for generating, evaluating, and visualizing enhanced pseudo masks for Weakly Supervised Semantic Segmentation (WSSS) using the Segment Anything Model (SAM).

• Host: GitHub
• URL: https://github.com/cskyl/SAM_WSSS
• Owner: cskyl
• Created: 2023-05-06T21:23:44.000Z (over 1 year ago)
• Default Branch: main
• Last Pushed: 2023-10-09T15:43:30.000Z (over 1 year ago)
• Last Synced: 2023-10-09T17:09:18.716Z (over 1 year ago)
• Topics: pascal-voc, sam, weakly-supervised-learning, wsss
• Language: Python
• Homepage:
• Size: 3.51 MB
• Stars: 31
• Watchers: 3
• Forks: 0
• Open Issues: 6
• Metadata Files:
  • Readme: README.md

Awesome Lists containing this project

• Awesome-Segment-Anything - [code

README

        
# SAM_WSSS

Code repository for our paper "[Segment Anything Model (SAM) Enhanced Pseudo

Labels for Weakly Supervised Semantic Segmentation](https://arxiv.org/abs/2305.05803)"

This is a Python script for our proposed framework.

## Getting Started

These instructions will get you a copy of the project up and running on your local machine.

### Prerequisites

- Python 3.6 or higher

### Installation

Clone the repository:

```bash

git https://github.com/cskyl/SAM_WSSS.git

cd SAM_WSSS

```

### Running mask enhancement

You can specify the locations of your pseudo labels and SAM masks folders by modifying the pseudo_path and sam_path respectively. Use the command below, replacing your_pseudo_labels and your_SAM_masks with the corresponding paths on your system:

```

python main.py --pseudo_path  --sam_path  

```

This will use our default merging algorithm, and the result will be stored in 

```

SAM_WSSS/processed_masks

```

### Evaluation and Visualization

To evaluate the quality of the pseudo masks and enhanced masks, you can use the following command. Please note that the current version of our code supports evaluation for object classes from the PASCAL VOC 2012 dataset.

```

python main.py --mode eval --pseudo_path  --sam_path  --gt_dir 

```

This script will generate evaluation statistics for each image sample, which will be saved in a .csv file in the following directory:

```

SAM_WSSS/eval

```

For visualization purposes, you can use the command below. You have the option to visualize specific samples by using vis_sample. Alternatively, you can use vis_best or vis_worst to visualize the top 10 samples that showed the most improvement or the least improvement, respectively, in delta mIoU after enhancement.

```

python main.py --mode vis --pseudo_path  --sam_path  --gt_dir  --images_path 

```

This will generat the visualization in

```

SAM_WSSS/visualizations

```

## Customizing the Merging Algorithm

If you wish to implement your own merging algorithm, you can modify the `merge` function located in the following script:

```

SAM_WSSS/merge/merge_customize

```

To run your custom algorithm, use the following command:

```

python main.py --pseudo_path  --sam_path  --method 'merge_customize'

```

By changing the --mode argument to 'all', you can run the entire pipeline, which includes merging, evaluation, and visualization. 

If you are using our code, please consider citing our paper.

```

@misc{chen2023segment,

      title={Segment Anything Model (SAM) Enhanced Pseudo Labels for Weakly Supervised Semantic Segmentation}, 

      author={Tianle Chen and Zheda Mai and Ruiwen Li and Wei-lun Chao},

      year={2023},

      eprint={2305.05803},

      archivePrefix={arXiv},

      primaryClass={cs.CV}

}

```