Improve model card and surface related exports

9723c01 verified about 23 hours ago

2.08 kB

---
license: agpl-3.0
pipeline_tag: object-detection
tags:
- rknn
- rknn2
- rockchip
- yolov10
- rk3576
- rk3588
---
# YOLOv10-RKNN2-Ultralytics

> **Looking for the Rockchip Model Zoo and custom exports?** See [YOLOv10-RKNN2](https://huggingface.co/HanzoHuang/YOLOv10-RKNN2).

RKNN exports of Ultralytics YOLOv10 for **RK3576** and **RK3588**, compiled with RKNN Toolkit2 2.3.2. Detection variants n, s, m, b, l, and x are available in FP16 and INT8.

## Choose a model

Models are organized by toolkit version, task, variant, target chip, and precision:

```text
models/v<toolkit-version>/<task>/<variant>/<chip>/<precision>/
```

Download `model.rknn` and `metadata.yaml` from the same directory. The [manifest](manifest.json) lists the available models and checksums.

## INT8 detection

Use the `metadata.yaml` from the same model directory. INT8 detection outputs normalized box coordinates and requires `coordinates: normalized`. Import both files together in the warehouse, or keep them in the same directory for the CLI. Treating normalized coordinates as pixels can result in no boxes.

## Reproduce the conversion

The [source directory](tree/main/source) contains the input checkpoints, conversion scripts, package versions, and calibration hashes. The export used [Ultralytics 8.4.184](https://pypi.org/project/ultralytics/8.4.184/#files) and [RKNN Toolkit2 2.3.2](https://github.com/airockchip/rknn-toolkit2/tree/v2.3.2).

```bash
python source/convert-model.py source/weights/yolov10n.pt --chip rk3588 --precision fp16 --imgsz 640 --output yolov10n-rk3588
```

For INT8, use `--precision int8 --data /path/to/coco128.yaml --calibration-limit 128`. Point the YAML at the extracted COCO128 tree and its `train2017` image directory. The calibration manifest links to the source images.

## Source and license

The inputs are official Ultralytics checkpoints. Conversion changes the format and quantizes the INT8 graph. The RKNN exports and original checkpoints are distributed under AGPL-3.0; see [LICENSE](LICENSE) and [checkpoint origins](source/weights.json).

	---
	license: agpl-3.0
	pipeline_tag: object-detection
	tags:
	- rknn
	- rknn2
	- rockchip
	- yolov10
	- rk3576
	- rk3588
	---

	# YOLOv10-RKNN2-Ultralytics

	> Looking for the Rockchip Model Zoo and custom exports? See [YOLOv10-RKNN2](https://huggingface.co/HanzoHuang/YOLOv10-RKNN2).

	RKNN exports of Ultralytics YOLOv10 for RK3576 and RK3588, compiled with RKNN Toolkit2 2.3.2. Detection variants n, s, m, b, l, and x are available in FP16 and INT8.

	## Choose a model

	Models are organized by toolkit version, task, variant, target chip, and precision:

	```text
	models/v<toolkit-version>/<task>/<variant>/<chip>/<precision>/
	```

	Download `model.rknn` and `metadata.yaml` from the same directory. The [manifest](manifest.json) lists the available models and checksums.

	## INT8 detection

	Use the `metadata.yaml` from the same model directory. INT8 detection outputs normalized box coordinates and requires `coordinates: normalized`. Import both files together in the warehouse, or keep them in the same directory for the CLI. Treating normalized coordinates as pixels can result in no boxes.

	## Reproduce the conversion

	The [source directory](tree/main/source) contains the input checkpoints, conversion scripts, package versions, and calibration hashes. The export used [Ultralytics 8.4.184](https://pypi.org/project/ultralytics/8.4.184/#files) and [RKNN Toolkit2 2.3.2](https://github.com/airockchip/rknn-toolkit2/tree/v2.3.2).

	```bash
	python source/convert-model.py source/weights/yolov10n.pt --chip rk3588 --precision fp16 --imgsz 640 --output yolov10n-rk3588
	```

	For INT8, use `--precision int8 --data /path/to/coco128.yaml --calibration-limit 128`. Point the YAML at the extracted COCO128 tree and its `train2017` image directory. The calibration manifest links to the source images.

	## Source and license

	The inputs are official Ultralytics checkpoints. Conversion changes the format and quantizes the INT8 graph. The RKNN exports and original checkpoints are distributed under AGPL-3.0; see [LICENSE](LICENSE) and [checkpoint origins](source/weights.json).