RapidOCR PP-OCRv5 ONNX (det + rec + dict)

Serving repo for the PaddleOCR PP-OCRv5 mobile detection + recognition pair (DBNet-style det probability map, SVTR-HGNet rec, 18,385 classes) plus the ppocrv5_dict.txt charset the ONNX exports don't carry.

Provenance

  • ppocrv5_det_infer.onnx / ppocrv5_rec_infer.onnx: byte-identical to PaddlePaddle/PP-OCRv5_mobile_det_onnx / PP-OCRv5_mobile_rec_onnx inference.onnx (Apache-2.0), renamed to flat root names: det sha256:a431985659dc921974177a95adcfbb90fd9e51989a5e04d70d0b75f597b6e61d, rec sha256:da72dc72ca4dc220df0dfde68c1dedc31c58d3e76a25871122e5056227d50092.
  • ppocrv5_dict.txt (18,383 lines): PaddleOCR ppocr/utils/dict/ppocrv5_dict.txt verbatim. CTC classes = 18,383 chars + blank (index 0) + space (last).

Serving (bobine OCR slot)

Det input x ([1,3,960,960] ImageNet canvas) β†’ single fetch_name_0 probability map (DB-decode in caller). Rec input x ([1,3,48,W]) β†’ fetch_name_0 [B,T,18385] logits, greedy CTC with the dict above.

Downloads last month

-

Downloads are not tracked for this model. How to track
Inference Providers NEW
This model isn't deployed by any Inference Provider. πŸ™‹ Ask for provider support