4.0.1

Changelog

1. Reduced code duplication in Github Actions Workflow files.
2. Added CI/CD to automatically publish the carvekit package to the PyPi package registry.
3. Carve Kit package for Google Colab published on PyPI.
4. Simplified framework installation procedure
5. Added console alias carvekit to interact with the console interface without python3 -m carvekit.
6. SSRF filtered in API endpoint

Links to pacakges:

• https://pypi.org/project/carvekit/
• https://pypi.org/project/carvekit-colab/

4.0

Changes:

2. Optimized the fba post-processing method for basnet, deeplabv3 models.
3. The deprecated deeplabv3 tensorflow models were removed instead of them one deeplabv3 model was added with the resnet101 backbone from the pytorch hub. Also now the project does not depend on tensorflow.
4. Improved trimap generators.
5. Removed deprecated tests. Also added testing of the program on different platforms (Windows, MacOS, Linux).
6. Added automatic creation of a Docker container after testing program and testing tool in it.
7. Fixed serious bugs in Flask/FastAPI http api. In particular, a memory overflow error with a large number of requests, an error with the processing of some parameters in a request when removing the background, an error in processing a system variable when working in a Docker container, etc.
8. Added a queue for requests in the FastAPI http api, also added a simple implementation of the Task Queue system with a time for deleting unclaimed server responses in 1 hour after processing.
9. All links, disclaimers have been added in CREDITS.md.
10. A large-scale work has been done with README.md. Added new example images, also written new examples for working with the FastAPI http api, updated information about building a docker container, updated information about the arguments of the console interface, updated information about installing the program, etc.
11. Updated project dependencies.
12. Completely rewritten code of framework.
13. The consumption of video memory and RAM is maximally optimized.
14. Optimized the code of post-processing methods. #29
15. Now the console interface accepts image paths as the user expects. #26
16. Also added a recursive search for images in a folder.
17. Dropped python 3.5 support.
18. Updated colab notebook.
19. Refactored Docker
20. Added separate Docker images building for CPU and GPU processing.
21. Models downloader now inside framework and downloads all automaticly to ~/.carvekit.
22. Refactored and rewrited the HTTP API to the FastAPI backend.
23. Updated and optimized post-processing methods and segmentation models to improve quality.
24. Added simple front-end to HTTP API.
25. Refactored demo code
26. Updated Github CI/CD workflows ymls
27. Added docker-compose files
28. Added good support for NVIDIA CUDA processing devices.
29. Added 100% test coverage of the main code.

What's Changed

• Add docker build, Improve http_api by @pubkey in #25
• Loading models via absolute path by @chris-rgr in #28
• Bump tensorflow from 2.2.0 to 2.2.1 by @dependabot in #35
• Bump pillow from 7.1.2 to 8.1.1 by @dependabot in #49
• Fix warning about mixed memory format pytorch by @xhit in #52
• First stage of refactor by @MrFox131 in #69
• [4.0] Complete refactor by @OPHoperHPO in #70
• [ALPHA][Version 3.3] -> [RELEASE][Version 4.0] by @OPHoperHPO in #33
• [RELEASE][Version 4.0] Complete Refactor by @OPHoperHPO in #22
• [RELEASE][4.0] by @OPHoperHPO in #71

New Contributors

• @pubkey made their first contribution in #25
• @chris-rgr made their first contribution in #28
• @dependabot made their first contribution in #35
• @xhit made their first contribution in #52
• @MrFox131 made their first contribution in #69

Full Changelog: 3.2.1...4.0

3.2.1

Changes:

1. Changed models links in the setup.py

Release 3.2! Added methods of preprocessing and post-processing of images!

Changes:

1. Updated README.md
2. Added support for the new neural network: BASNet
3. Added tests for the program.
4. MODELS_NAMES moved to strings.py.
5. Added CREDITS.md
6. Updated sample photos
7. Add gui #10
8. Added methods of preprocessing and post-processing of images that improve the quality of removing background from photos
9. Rewritten libs/networks.py
10. Fixed memory leak in gui.py
11. Fixed #19
12. And many other minor changes.

Upd 24.06.2020:
The following files are pre-trained models of neural networks and do not require downloading. They are here because the original links to these models from the authors of neural networks have many restrictions on the loading of models. These models will be automatically downloaded and installed on your computer using the setup.sh script.

Release 3.1! Added support for new neural network!

Changes

1. The code base is completely rewritten
2. Added support for a new neural network (U^2-Net)
3. Added new models to the installer script
4. Added selective installation of models in the installer script
5. Improved manual
6. Updated old code
7. Added MODELS.md with information about used neural networks
8. Added LICENSE
9. Updated examples .shs and .bats
10. Added some libs.
11. Changes cli args.
12. And many other minor changes and fixes.

Release v3.0! Many improvements!

Changelog

• The code base is completely rewritten.
• Changed model storage paths.
• Rewritten installation script from bash to python language.
• Added full script compatibility with Windows (NOT TESTED).
• Improved manual.
• Added examples of using the program.
• Added versions of the necessary dependencies for the program (Added requirements.txt).
• Tensorflow 2.0 compatible.
• The script now not only processes a single file, but can also process all images from the input folder and save them in the output folder with the same name.
• And many other minor changes and fixes.

2.0.6 [Public] - New features!

Changelog:

• Added comments to the code.
• Added tqdm progress bar.
• Removes background from image without loss of image resolution.
• The script now processes all images from the input folder and saves them to the output folder with the same name.
• New sample images.