Views
No views yet
torchvision::deform_conv2d (deformable convolution), which neither the legacy nor
the dynamo ONNX exporter supports. This export replaces that op with a mathematically equivalent
implementation built from grid_sample (validated numerically against torchvision.ops.deform_conv2d
to < 1e-3), so the model runs in ONNX Runtime / Core ML EP.| Input | input_image, float32, shape [1, 3, 1024, 1024], NCHW |
| Normalization | ImageNet — mean [0.485, 0.456, 0.406], std [0.229, 0.224, 0.225] |
| Output | output_image, [1, 1, 1024, 1024] logits — apply sigmoid for the alpha matte |
| Precision | fp32 (~897 MB) · opset 16 |
ZhengPeng7/BiRefNet-matting.1@article{BiRefNet,
2 title={Bilateral Reference for High-Resolution Dichotomous Image Segmentation},
3 author={Zheng, Peng and Gao, Dehong and Fan, Deng-Ping and Liu, Li and Laaksonen, Jorma and Ouyang, Wanli and Sebe, Nicu},
4 journal={CAAI Artificial Intelligence Research},
5 year={2024}
6}