A package for clip-guided diffusion
Project description
SAM + CLIP + DIFFUSION
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.
Installation
pip install samclipdiffusion
Usage
!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
Examples with and w/o installing the package.
Citation
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
TODO
- Add more examples
- Optimize, make it memory efficient, it's awful right now
- Add better images to the examples :D
Project details
Release history Release notifications | RSS feed
Download files
Download the file for your platform. If you're not sure which to choose, learn more about installing packages.
Source Distribution
Built Distribution
File details
Details for the file samclipdiffusion-0.0.1b2.tar.gz
.
File metadata
- Download URL: samclipdiffusion-0.0.1b2.tar.gz
- Upload date:
- Size: 4.3 kB
- Tags: Source
- Uploaded using Trusted Publishing? No
- Uploaded via: twine/4.0.2 CPython/3.8.3
File hashes
Algorithm | Hash digest | |
---|---|---|
SHA256 | e2647deabe04a4006dc061ab99a9cb47a6a32f39c9889fffe0a3d19b3b7a79a7 |
|
MD5 | 3bdb1102f2be2ec9a582a51ec66685de |
|
BLAKE2b-256 | 8a8aeb5c0a49a4bd47b05e1164eefe0534ba316bda81f25fff1089b23e8646c0 |
File details
Details for the file samclipdiffusion-0.0.1b2-py3-none-any.whl
.
File metadata
- Download URL: samclipdiffusion-0.0.1b2-py3-none-any.whl
- Upload date:
- Size: 4.6 kB
- Tags: Python 3
- Uploaded using Trusted Publishing? No
- Uploaded via: twine/4.0.2 CPython/3.8.3
File hashes
Algorithm | Hash digest | |
---|---|---|
SHA256 | c10f0dff36f538aa7ef33c0e69ed3d94deb12f9cdcf093784b62776ddead9106 |
|
MD5 | 64a33dfd1c6f1939bf035881eb563a61 |
|
BLAKE2b-256 | 67128f9de53e8f4101484d1644709ff900d82277cace6f9f76bfe962c709dcd3 |