Run ECCV2022-RIFE on AMD ROCM
Loganavter opened this issue · 1 comments
I created a venv in the ECCV2022-LIFE folder, started the
pip3 install -r installation requirements.txt
and then I installed my torch, torchaudio and torchvision in this venv in order for rocm to start working on my RX 6600 video card
pip install torch==2.0.1+rocm5.4.2 torchvision==0.15.2+rocm5.4.2 torchaudio==2.0.2+rocm5.4.2 -f https://download.pytorch.org/whl/rocm5.4.2/torch_stable.html After that, I downloaded a demo video from readme.md and rife_trained_v6 tried to do frame interpolation,
(venv) [jorj@jorj-systemproductname ECCV2022-RIFE]$ python3 inference_video.py --exp=1 --video=demo.mp4
Loaded ArXiv-RIFE model
demo.mp4, 1342.0 frames in total, 25.0FPS to 50.0FPS
The audio will be merged after interpolation process
0%| | 0/1342.0 [00:00<?, ?it/s]Segmentation error (the memory stack is flushed to disk)
As you can see above, I get an error when trying to use it, can you help me fix the error? I had no problems using rocm with stable diffusion a1111, ComfyUI and ChaiNNer.
Bruh, i forgot to use HSA_OVERRIDE_GFX_VERSION=10.3.0 LD_PRELOAD=libtcmalloc.so.4