MAT Places 512 for the LiteRT GPU delegate

MAT (Mask-Aware Transformer, Li et al., CVPR 2022) Places 512 inpainting model, rewritten so every operation runs on the LiteRT (TensorFlow Lite) GPU delegate on Android.

  • Inputs: RGB image 1x512x512x3 in 0..1 and erase mask 1x512x512x1 (1 = fill).
  • Output: RGB 1x512x512x3 in 0..1.
  • Weights are FP32; run with precision loss allowed (FP16) for speed.

Converted from the FP16 weights at https://huggingface.co/spacepxl/MAT-inpainting-fp16 (original: https://github.com/fenglinglwb/MAT).

License

The MAT weights are licensed under CC BY-NC 4.0, so this derived model is for non-commercial use only. Credit: Wenbo Li, Zhe Lin, Kun Zhou, Lu Qi, Yi Wang, Jiaya Jia, "MAT: Mask-Aware Transformer for Large Hole Image Inpainting", CVPR 2022.

Downloads last month
-
Inference Providers NEW
This model isn't deployed by any Inference Provider. ๐Ÿ™‹ Ask for provider support