Kneron Model Zoo Inference Example - Detection(Yolov5)
PLUS Inference Example with NEF model trained from https://doc.kneron.com/docs/#model_training/object_detection_yolov5/
Check - tutorial ( location: /workspace/ai_training/detection/yolov5/tutorial/tutorial.ipynb )
in
- Kneron Toolchain docker (doc: https://doc.kneron.com/docs/#toolchain/manual_1_overview/)
to learn how to train a the example model, and check
- https://doc.kneron.com/docs/#toolchain/manual_1_overview/
- https://doc.kneron.com/docs/#toolchain/appendix/yolo_example/
to learn how to convert the trained model to NEF model
Note: We built and run the examples below under OS Windows 10 (19041.1052), Keil uVision5 (5.27.1.0), MSYS2 MinGW 64-bit (20210419), and cmake version 3.20.2.
How to:
step 1. read document at Getting_start and make sure PLUS is ready
step 2-1. run KL520 example:
cd build/bin/
./kl520_kn-model-zoo_generic_inference_post_yolov5 step 2-2. run KL720 example:
cd build/bin/
./kl720_kn-model-zoo_generic_inference_post_yolov5 Detail Infomation:
| Platform | backbone | FPS(npu only) | Input format | Preprocess | Normalize |
|---|---|---|---|---|---|
| KL520 | YOLOv5s | 4.91429 | RGB565 | auto (on device) | Kneron mode |
| KL720 | YOLOv5s | 24.4114 | RGB565 | auto (on device) | Kneron mode |