Skip to content

CHANGELOG.md

Latest commit

 

History

History
25 lines (18 loc) · 1.09 KB

CHANGELOG.md

File metadata and controls

25 lines (18 loc) · 1.09 KB
Raw

Changelog

[1.0.0] - Initial Release

Changed

  • inference_realesrgan_video.py

    • Added a compatibility patch for torchvision.
    • Work in progress to align with the image inference path, including device handling, input validation, and logging.

  • inference_realesrgan.py

    • Added a compatibility patch for torchvision.
    • Introduced a --device argument to specify gpu or cpu (default: gpu).
    • Enhanced GPU selection logic to auto-select the most capable GPU or use a specific GPU via --gpu-id.
    • Improved input filtering to process only valid image files.

  • requirements.txt

    • Updated dependencies for compatibility with recent PyTorch and CUDA stacks, targeting CUDA 12.8 and Blackwell-class GPUs.

  • realesrgan/utils.py

    • Added a _get_most_capable_gpu() function to rank GPUs by compute capability and memory.
    • Enhanced device selection logic to prioritize the best GPU or fallback to CPU when explicitly requested.

  • torchvision_patch.py

    • Introduced a compatibility patch to resolve import issues with torchvision.transforms.functional_tensor.