zhaozhengchen / sam_wsss

SAM_WSSS

SAM for Weakly-Supervised Semantic Segmentation (WSSS).

Environment

Our code is based on Grounded-Segment-Anything. Please refer to it on installation.

Usage

PASCAL VOC

SAM (text input)

python sam_text_input_voc.py --img_root YOUR_DATA_PATH --mask_dir YOUR_OUTPUT_PATH

SAM (tero-shot)

python sam_zero_shot_voc.py --img_root YOUR_DATA_PATH --mask_dir YOUR_OUTPUT_PATH

MS COCO

SAM (text input)

python sam_text_input_coco.py --img_root YOUR_DATA_PATH --mask_dir YOUR_OUTPUT_PATH

SAM (tero-shot)

python sam_zero_shot_coco.py --img_root YOUR_DATA_PATH --mask_dir YOUR_OUTPUT_PATH

Acknowledgment

Thanks to Grounded-Segment-Anything.

sam_wsss's Issues

Why this repo so good?

Please justify why this repo is so good? Here we define the "repo" as a virtual space containing programming codes held by GitHub servers. Then how to scientifically measure the goodness/wellness of a GitHub repo? What forms a scientifical/critical measurement? We propose to measure this goodness from two dimensions, which is:

1. I don't know
2. I don't know

This measurement has already been validated by the community and is rooted from "the Stone and Pig theory". Based on this criteria, we show that this SAM_WSSS repo is scientifcally good:

1. Because I don't know, so it is really good.
2. The same reason as in 1.

Given this, we validated that this proposed SAM_WSSS is quite good repo. #