Lin Song
Lin Song
Thanks for your interest in our work. 1) Because out_data is ordered as the input feature, but out_grad is ordered in the breadth-first-search (BFS) manner. 2) The "weight" in "refine.cu"...
I have not tried this specific case, but I think it will work fine. It is easy to implement in your codebase. You can add a tree filter after each...
Yes, it can run on V100 GPU. What issues did you have when compiling the code?
I have updated the code. Now it can support cuda 11 and up to pytorch 1.12.1.
Can you provide more compiling details? such as warnings or errors.
Since the tree filter kernel relies on some components of the PyTorch, please executing "import torch" before "import tree_filter_cuda".
This could be caused by the inconsistent version of PyTorch between compiling and runtime. You can try to reinstall PyTorch and refresh the environment, or rerun the installation procedure. In...
Please provide more details. It seems to be the issues of environment variables of CUDA libraries. You can refer to this solution [https://github.com/pytorch/fairseq/issues/6#issuecomment-332351111](https://github.com/pytorch/fairseq/issues/6#issuecomment-332351111).
This file is generated by compiling the cuda codes, that could be different for different machines. You can refer to the building guide at [https://github.com/StevenGrove/vtpack#build-from-source](https://github.com/StevenGrove/vtpack#build-from-source)
Please provide more detailed logs or error messages.