jamesjg
commented
5 months ago Hi, thanks for your time!
We directly clone mmseg v0.11.0 in order to match the version used by FoodSeg103 since we leverage the baseline model SETR from FoodSeg103 for coarse segmentation. So far we have not tested FoodSAM with newer versions of mmseg or mmcv. Sticking with the matched versions avoids environment configuration issues. Upgrading would require careful testing for compatibility, which we have not prioritized yet.
If you use other segmentation models, you may need use its corresponding version. Thanks again for your interest in FoodSAM!
Hello,
First of all, thank you very much for your hard work and for sharing this project with the community! I've been exploring FoodSAM and am excited about the potential it has. However, I have a few questions regarding its dependencies and compatibility with newer versions of certain libraries. Your insights on these would greatly help me integrate and potentially contribute to the project:
Integration of mmseg v0.11.0: I noticed that the project directly clones the mmseg v0.11.0 code. Could you share the reasoning behind this specific version choice? Are there specific features or compatibility issues with newer versions that influenced this decision? Did you make important changes in the code and using it on FoodSAM ?
Compatibility with Newer Versions of mmseg: Have there been any attempts to integrate newer versions of mmseg into FoodSAM? If so, were there any notable challenges or findings from those attempts?
Use of Older mmcv Versions: The project currently utilizes mmcv-full 1.3.0. Have there been any tests or considerations for upgrading to more recent versions, such as mmcv-full 1.7 or even mmcv 2? If compatibility issues arose, could you detail them?
Understanding these aspects would not only help me in utilizing FoodSAM more effectively but also in potentially contributing to its development, ensuring compatibility and leveraging improvements from these libraries.
Thank you for your time and assistance!
Best regards, Jules