Stuff about this pipeline was a bit harder to run and follow, I wanted to write a simple package to allow people to use it with a higher level of abstraction. Based on the repos on the citation part.
pip install samclipdiffusion
!wget https://dl.fbaipublicfiles.com/segment_anything/sam_vit_h_4b8939.pth
from samclipdiffusion import ImageSegmenter
segmenter_inp = ImageSegmenter()
image_path = 'image_path'
search_text = "target object"
prompt = "how to modify"
inpainted_image = segmenter_inp.inpaint_image(image_path, search_text, prompt)
Examples with and w/o installing the package.
This repository is based on the following repos, I just merged them together and made some changes to make it work.
https://github.com/IDEA-Research/Grounded-Segment-Anything/blob/main/grounded_sam.ipynb https://github.com/maxi-w/CLIP-SAM/blob/main/main.ipynb
- Add more examples
- Optimize, make it memory efficient, it's awful right now
- Add better images to the examples :D