Hi, thank you for releasing this project! 🙏 I’ve been trying to set up the environment following the provided installation.md, but I’ve run into some version conflicts and would appreciate clarification.

Specifically, could you help clarify the following points?

1. Exact package versions

• The requirements.txt lists only package names (e.g., opencv-python, transformers, timm), but many of these have breaking changes across versions.
• Could you please share a pip list --format=freeze or conda env export from the environment you used to run your experiments? That would make it much easier to reproduce the setup.

2. Non-pip dependencies

• Some modules (e.g., libs/pointops, spconv, flash-attn, tiny-cuda-nn) need to be compiled or installed in special ways.
• Could you clarify which ones require source builds and which ones can be installed directly via pip/conda?

3. Hardware / CUDA dependency

• The README mentions testing on a 24 GB RTX 4090 (Ada architecture).
• Is the code known to work on other GPUs (e.g., 30-series or 50-series), or are there assumptions tied to the 40-series hardware/CUDA arch?

Thanks a lot for your help, and for sharing the code! 🙌 I think this clarification would also help many other users avoid version mismatches and make the repo more accessible.