This is a implementation of the Differentiable Binarization. And this model perform the binarization process in a segmentation network. We directly adapt the PaddleOCR version. And we adapt MobileNet-v2 as the backbone.
| Model(nntc) | Model(dbnet) | H1 Mean |
|---|---|---|
| dbnet.pdmodel | dbnet.onnx | 68% |
You can use Paddle2Onnx to convert paddle-style model-storage to onnx-style.
We choose Task 4.1 in Incidental Scene Text Dataset icdar2015 as the benchmark. And we offer preprocessed data and groundtruth.
Apache-2.0