BiRefNet_lite 512 (WebGPU)

BiRefNet_lite (MIT, by ZhengPeng7) at 512x512, fp16 ONNX, from studioludens/birefnet-lite-512. The graph is rewritten so it runs fully on WebGPU within phone GPU limits: deformable-conv GatherND nodes and wide Split nodes are replaced with equivalent ops using the MIT patch scripts from jiabins0303/birefnet-lite-1024-webgpu. Output is bit-identical to the original.

Downloads last month
66
Inference Providers NEW
This model isn't deployed by any Inference Provider. 🙋 Ask for provider support

Model tree for Wattsey/birefnet-lite-512-webgpu

Quantized
(14)
this model