search

hzwer / ECCV2022-RIFE

ECCV2022 - Real-Time Intermediate Flow Estimation for Video Frame Interpolation
MIT License
4.39k stars 438 forks source link

Run ECCV2022-RIFE on AMD ROCM #354

Closed Loganavter closed 8 months ago

Loganavter commented 8 months ago

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.

Loganavter commented 8 months ago

Bruh, i forgot to use HSA_OVERRIDE_GFX_VERSION=10.3.0 LD_PRELOAD=libtcmalloc.so.4