Error converting from onnx to ncnn about ncnn-android-yolov7 HOT 3 OPEN

you have to exclude the detect layer while exporting from pytorch to onnx, you are trying to export end-to-end model.

from ncnn-android-yolov7.

Hi xiang-wu,

Thanks for your quick feedback. I've corrected my error, but I'm still having problems with the model.
After exporting the custom weights to nccn and running the application, I realized that net doesn't do any detection (the anchor boxes are never shown), the application works like a camera.
Below is the step by step that I am using.

a) My network is used to detect fruit diseases. It comprises about 20,000 images, with 9 classes and achieves a mAP of approximately 87% in yolov7-tiny

b) .PT --> onnx
python export.py --weights yolo-papaya-tiny.pt --simplify --topk-all 100 --iou-thres 0.50 --conf-thres 0.15 --img-size 448 448 --max-wh 448

c) onnx --> ncnn

./onnx2ncnn yolo-papaya-tiny.onnx yolo-papaya-tiny.param yolo-papaya-tiny.bin

d) After conversion, the weights are renamed and transferred to the application folder

If you have the time and inclination to point me out where I'm going wrong, I'd be very grateful.
I am also attaching the weights after converting to ncnn

ncnn - yolo-papaya-tiny.zip
Origin-yolo-papaya-tiny.zip

from ncnn-android-yolov7.

you have to exclude the detect layer while exporting from pytorch to onnx, you are trying to export end-to-end model.

能加个邮箱吗？我们有些问题想要咨询，我的邮箱在主页介绍那里

from ncnn-android-yolov7.