Compatibility with MMCV 2.2.0 and PyTorch 2.3 Development Plan

[specter119]

Dear mmlab team,

I am writing to inquire about the compatibility of mmdetection with the latest version of mmcv (2.2.0) and its support for PyTorch 2.3. This concern is heightened by the recent security vulnerability reported as CVE-2024-5480 (https://www.tenable.com/cve/CVE-2024-5480).

The CVE report indicates a critical security issue resolved in versions of PyTorch prior to 2.2.2. Consequently, it is imperative for users to upgrade to a secure version to protect their systems. However, the current release of mmdetection does not support mmcv 2.2.0, which is necessary for compatibility with PyTorch 2.3.

I have noticed an ongoing discussion and a pull request (PR #11764) that proposes loosening the version constraints for mmcv. This PR appears to be a positive step towards addressing the compatibility issue, but it does not completely resolve the problem.

Could the team please share any updates on the timeline or development roadmap for making mmdetection compatible with mmcv 2.2.0 and PyTorch 2.3? The resolution of this issue is crucial for the security and integrity of the numerous projects that depend on mmdetection.

I appreciate your commitment to this project and thank you in advance for your attention to this pressing security matter.

[crbean]

I had the same problem
Hope to provide a plan for adapting to mmcv 2.2.0.

[haridimos9]

I had the same problem too. I think (PR #11764) is a step towards enabling projects to continue working.
I noticed that the version error is not resolved by installing mmcv-lite 2.1.0.

[crbean]

@hhaAndroid This issue has lasted for one month. Is there any plan to release a version to solve the above problems?