card: update from fastforks 1cbf1671
Browse files
README.md
ADDED
|
@@ -0,0 +1,109 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 1 |
+
---
|
| 2 |
+
library_name: "fastforks"
|
| 3 |
+
pipeline_tag: "image-to-text"
|
| 4 |
+
license: "apache-2.0"
|
| 5 |
+
tags:
|
| 6 |
+
- "ocr"
|
| 7 |
+
- "text-recognition"
|
| 8 |
+
- "text-detection"
|
| 9 |
+
- "cuda"
|
| 10 |
+
- "pytorch"
|
| 11 |
+
- "fastforks"
|
| 12 |
+
- "fastforks-card"
|
| 13 |
+
base_model:
|
| 14 |
+
- "PaddlePaddle/PP-OCRv5_mobile_det"
|
| 15 |
+
- "PaddlePaddle/PP-OCRv5_mobile_rec"
|
| 16 |
+
- "PaddlePaddle/PP-OCRv5_server_det"
|
| 17 |
+
- "PaddlePaddle/PP-OCRv5_server_rec"
|
| 18 |
+
---
|
| 19 |
+
# PP-OCRv5 (server / mobile) (fastforks)
|
| 20 |
+
|
| 21 |
+
PP-OCRv5 (PaddleOCR 3, Baidu 2025): a DB text detector (PP-HGNetV2 + LKPAN for server, PP-LCNetV3 + RSE-FPN for mobile) finds text lines of any orientation as quads; each line is cut out with a perspective warp and read by a CTC recogniser (PP-HGNetV2 / PP-LCNetV3 with an SVTR neck) over 18 383 characters (simplified and traditional Chinese, English, Japanese, pinyin). Inference as PaddleX's OCR pipeline without the document unwarping and the text-line orientation classifier. The portable `module()` is the detector with its normalisation.
|
| 22 |
+
|
| 23 |
+
This repo is the model card only: it holds no weights. fastforks downloads the authors' weights from the links below and checks them against the SHA-256 in the model card. Code: <https://github.com/kleinicke/fastforks>. Kernels for the `kernels` backend: <https://huggingface.co/fastforks/kernels>. Original code: <https://github.com/PaddlePaddle/PaddleOCR> at `dab3fe353790`.
|
| 24 |
+
|
| 25 |
+
Author weight repos: <https://huggingface.co/PaddlePaddle/PP-OCRv5_mobile_det>, <https://huggingface.co/PaddlePaddle/PP-OCRv5_mobile_rec>, <https://huggingface.co/PaddlePaddle/PP-OCRv5_server_det>, <https://huggingface.co/PaddlePaddle/PP-OCRv5_server_rec>.
|
| 26 |
+
|
| 27 |
+
Task: ocr. Tier: experimental. Outputs: `boxes`, `texts`, `scores`. Backends: `portable`.
|
| 28 |
+
|
| 29 |
+
Input: one RGB image. Output: the text lines in reading order: a quad `[N, 4, 2]` per line, its text and a confidence `[N]` (text detector + recogniser; recognisers alone read the whole image as one line).
|
| 30 |
+
|
| 31 |
+
## Variants
|
| 32 |
+
|
| 33 |
+
Weights are the authors' files (not hosted here). SHA-256 shortened to 12 hex digits; the full hashes are in [`card.toml`](https://github.com/kleinicke/fastforks/blob/main/src/fastforks/ocr/models/ppocrv5/card.toml).
|
| 34 |
+
|
| 35 |
+
| variant | weights (author link) | SHA-256 | weights licence | commercial use |
|
| 36 |
+
|---|---|---|---|---|
|
| 37 |
+
| `ppocrv5-mobile@1` | [PP-OCRv5_mobile_det/inference.json](https://huggingface.co/PaddlePaddle/PP-OCRv5_mobile_det/resolve/0d63e78e2b680928f6b1747d76a08db6e645efb7/inference.json)<br>[PP-OCRv5_mobile_det/inference.pdiparams](https://huggingface.co/PaddlePaddle/PP-OCRv5_mobile_det/resolve/0d63e78e2b680928f6b1747d76a08db6e645efb7/inference.pdiparams)<br>[PP-OCRv5_mobile_rec/inference.json](https://huggingface.co/PaddlePaddle/PP-OCRv5_mobile_rec/resolve/682f20538d8c086cb2128e5cfac775e6c4904e85/inference.json)<br>[PP-OCRv5_mobile_rec/inference.pdiparams](https://huggingface.co/PaddlePaddle/PP-OCRv5_mobile_rec/resolve/682f20538d8c086cb2128e5cfac775e6c4904e85/inference.pdiparams)<br>[PP-OCRv5_mobile_rec/config.json](https://huggingface.co/PaddlePaddle/PP-OCRv5_mobile_rec/resolve/682f20538d8c086cb2128e5cfac775e6c4904e85/config.json) | `05feef1acb00`<br>`afa1820cb16c`<br>`24587345250c`<br>`2460da908759`<br>`6df1587ae056` | Apache-2.0 | yes |
|
| 38 |
+
| `ppocrv5-server@1` | [PP-OCRv5_server_det/inference.json](https://huggingface.co/PaddlePaddle/PP-OCRv5_server_det/resolve/ca867c897ecbca8873081573a802ad70d499cb94/inference.json)<br>[PP-OCRv5_server_det/inference.pdiparams](https://huggingface.co/PaddlePaddle/PP-OCRv5_server_det/resolve/ca867c897ecbca8873081573a802ad70d499cb94/inference.pdiparams)<br>[PP-OCRv5_server_rec/inference.json](https://huggingface.co/PaddlePaddle/PP-OCRv5_server_rec/resolve/b26c3587fda8da3c8ec0ce357214b4d661ff1558/inference.json)<br>[PP-OCRv5_server_rec/inference.pdiparams](https://huggingface.co/PaddlePaddle/PP-OCRv5_server_rec/resolve/b26c3587fda8da3c8ec0ce357214b4d661ff1558/inference.pdiparams)<br>[PP-OCRv5_server_rec/config.json](https://huggingface.co/PaddlePaddle/PP-OCRv5_server_rec/resolve/b26c3587fda8da3c8ec0ce357214b4d661ff1558/config.json) | `af5876933d88`<br>`183146fe9d99`<br>`8e6e12e5d425`<br>`63853f062a5f`<br>`6a780e213d7a` | Apache-2.0 | yes |
|
| 39 |
+
|
| 40 |
+
## Licence
|
| 41 |
+
|
| 42 |
+
Original-code strings, recorded as they are (research-use package; check them before any other use). Code: Apache-2.0. Licence files: `LICENSE`.
|
| 43 |
+
|
| 44 |
+
Training data: The PP-OCRv5 technical report describes public and synthetic text data plus Baidu's own data; the exact sets are not listed..
|
| 45 |
+
|
| 46 |
+
Licence evidence: https://github.com/PaddlePaddle/PaddleOCR/blob/dab3fe35379033fdcb2d0e9572fac0b36c9a9ebf/LICENSE ; https://huggingface.co/PaddlePaddle/PP-OCRv5_server_det
|
| 47 |
+
|
| 48 |
+
## Input size
|
| 49 |
+
|
| 50 |
+
any H x W >= 32, padded to a multiple of 32
|
| 51 |
+
Tested up to (portable, fp16, RTX 5060 Ti 16 GB): `ppocrv5-mobile` 3000x4000 (1.4 GiB peak allocated, 2.3 reserved); `ppocrv5-server` 3000x4000 (7.1 GiB peak allocated, 9.8 reserved)
|
| 52 |
+
|
| 53 |
+
## Speed and memory
|
| 54 |
+
|
| 55 |
+
### Ocr, synthetic
|
| 56 |
+
|
| 57 |
+
| model | input | time (ms): original fp32 | original | portable | peak memory (MiB, allocated / reserved): original | portable |
|
| 58 |
+
| --- | --- | --- | --- | --- | --- | --- |
|
| 59 |
+
| `ppocrv5-mobile` | 1080x1920 | 619 | 418 | 153 | 393 / 1005 | 289 / 448 |
|
| 60 |
+
| `ppocrv5-server` | 1080x1920 | β | 762 (fp32) | 277 | 2355 / 2901 | 1357 / 1452 |
|
| 61 |
+
|
| 62 |
+
### Startup and first call
|
| 63 |
+
|
| 64 |
+
| model | load (s): original | portable | kernels | first call (s): original | portable | kernels |
|
| 65 |
+
| --- | --- | --- | --- | --- | --- | --- |
|
| 66 |
+
| `ppocrv5-mobile` | 1.51 | 0.27 | β | 1.82 | 2.82 | β |
|
| 67 |
+
| `ppocrv5-server` | 2.07 | 0.56 | β | 2.03 | 1.87 | β |
|
| 68 |
+
|
| 69 |
+
### torch.compile and TensorRT
|
| 70 |
+
|
| 71 |
+
| model | input | time: portable | compile | TensorRT fp16 | kernels | preparation: compile (cold first call) | ONNX + engine build | peak MiB (alloc / reserved): portable | compile | kernels | TensorRT (engine + torch) |
|
| 72 |
+
| --- | --- | --- | --- | --- | --- | --- | --- | --- | --- | --- | --- |
|
| 73 |
+
| `ppocrv5-mobile` | 1080x1920 | 153 ms | 115 ms | 2.68 ms | β | 75.0 s | 156 s | 289 / 448 | 292 / 448 | β | 241 / 268 |
|
| 74 |
+
| `ppocrv5-server` | 1080x1920 | 277 ms | 277 ms | 25.1 ms | β | 105 s | 40.6 s | 1357 / 1452 | 958 / 1132 | β | 1131 / 1146 |
|
| 75 |
+
|
| 76 |
+
Fixed-size cells: time (ms) / peak allocated / peak reserved memory (MiB).
|
| 77 |
+
|
| 78 |
+
Hardware, protocol and caveats: [benchmarks](https://github.com/kleinicke/fastforks/blob/main/docs/benchmarks.md) (RTX 5060 Ti 16 GB, sm_120). Original = the original code named above at the precision of portable (fp16, or `(fp32)` where fp16 is unsupported or not finite); original fp32 = its default fp32 run. Peak memory is the torch allocator's.
|
| 79 |
+
|
| 80 |
+
## Accuracy
|
| 81 |
+
|
| 82 |
+
### Parity with the original code
|
| 83 |
+
|
| 84 |
+
Every variant is checked against reference outputs of the original code at its pinned commit. CPU, float32: max |diff| <= 0.001 vs the original. GPU, float16 (the default), portable and kernels: mean |diff| <= 0.5 and p99 <= 1.5 (absolute error of the output `boxes`).
|
| 85 |
+
|
| 86 |
+
## Limitations
|
| 87 |
+
|
| 88 |
+
- No document orientation / unwarping and no text-line orientation classifier (upside-down lines are read as they are).
|
| 89 |
+
- The DB post-processing runs on the CPU (OpenCV-exact geometry); hole contours of the binary map are not boxed (upstream boxes them too; they rarely pass `box_threshold`).
|
| 90 |
+
|
| 91 |
+
## Usage
|
| 92 |
+
|
| 93 |
+
```python
|
| 94 |
+
from fastforks import ocr
|
| 95 |
+
|
| 96 |
+
model = ocr.load("ppocrv5-mobile") # mode="auto": the preferred backend from recorded measurements
|
| 97 |
+
result = model("receipt.jpg")
|
| 98 |
+
result.texts # the text lines in reading order
|
| 99 |
+
result.boxes, result.scores # [N, 4, 2] quads in input pixels, [N] confidences
|
| 100 |
+
result.backend # what ran, e.g. Backend(mode='portable', device='cuda:0', dtype='float16')
|
| 101 |
+
```
|
| 102 |
+
|
| 103 |
+
Install with `pip install fastforks` (see <https://github.com/kleinicke/fastforks>). Other variants: `ppocrv5-mobile`, `ppocrv5-server`.
|
| 104 |
+
|
| 105 |
+
## Kernels backend
|
| 106 |
+
|
| 107 |
+
No kernels backend: cuDNN convolutions are the networks; the time outside them is the per-line crop / resize and the CPU box geometry.
|
| 108 |
+
|
| 109 |
+
<!-- staged from fastforks 1cbf1671 (2026-11-15) -->
|