Semantic-Fast-SAM

SSA + FastSAM
I'm inspired by FastSAM that higly faster than sam but sustains sam's performance succesfully.
(obviously, there are few Degraded performance, but still efficient i think)
this Semantic-Fast-SAM is also inspired by Semantic-Segment-Anything(i.e.,SSA). but change main segmentation branch, SAM(vit-h) to FastSAM(YOLOv8-seg).

Installation

To install the required dependencies, run the following command:

git clone https://github.com/KBH00/Semantic-Fast-SAM.git

cd Semantic-Fast-SAM

conda env create -f environment.yaml

conda activate sfs

python -m spacy download en_core_web_sm

You need to download Fast-SAM model checkpoint in here
Put FastSAM.pt in weights directory, mabye you should change the file name FastSAM-x.pt to FastSAM.pt

Inference

python scripts/main_ssa_engine.py --data_dir=data/<The name of your image> --out_dir=output --world_size=<choose GPU number>

Or, you can just run main_ssa_engine.py

Examples

References

Fast Segment Anything https://github.com/CASIA-IVA-Lab/FastSAM

Semantic Segment Anything https://github.com/fudan-zvg/Semantic-Segment-Anything