Release v3.9.0

Supplemental Notes for Update Log

1. This release sets the PP-OCRv6 det small detection model and rec small recognition model as the default models. Benchmark metrics demonstrate that PP-OCRv6 delivers significantly better performance than PP-OCRv4.
2. The total size of the whl package has increased from the previous 15 MB to 29 MB, resulting from the larger file sizes of the det small and rec small models.
3. This version currently supports inference with only two engines: ONNXRuntime and OpenVINO. Support for Paddle, Torch, MNN and TensorRT will be added incrementally in subsequent releases.

补充说明

1. 该版本将 PP-OCRv6 的 det small 模型和 rec small 识别模型作为了 默认模型。因为从指标来看，PP-OCRv6 效果远高于 PP-OCRv4。
2. 整体 whl 包从之前的 15 MB → 29M，因为 det small 和 rec small 模型变大了。
3. 该版本目前 仅支持了 ONNXRuntime 和 OpenVINO 两个推理引擎的推理，后续会逐步支持 Paddle, Torch, MNN 和 T ensorRT。

---

🚀 Features

• feat(python): support PP-OCRv6 (onnxruntime + openvino, det + rec) by @jaminmei in #692

🎉 Contributors

• @jaminmei made their first contribution in #692

Full Changelog: v3.8.4...v3.9.0