search

RangiLyu / nanodet

NanoDet-Plusāš”Super fast and lightweight anchor-free object detection model. šŸ”„Only 980 KB(int8) / 1.8MB (fp16) and run 97FPS on cellphonešŸ”„
Apache License 2.0
5.69k stars 1.03k forks source link

NCNN keeps reporting errors in reasoning #136

Open TD-wzw opened 3 years ago

RangiLyu commented 3 years ago

Can you post the error messageļ¼Ÿ

TD-wzw commented 3 years ago
error
RangiLyu commented 3 years ago

Are you using the coco pre-trained model from this repo or your own model? It seems that the index is out of range. Check out if the input_size in nanodet.h is same with your input image and the num_class is same with your model.

TD-wzw commented 3 years ago

OK, let me check. Thank you

TD-wzw commented 3 years ago

Check that this is not the problem, the outer loop can be executed 109 times

TD-wzw commented 3 years ago

*for (int idx = 0; idx < feature_h feature_w; idx++)** This sentence

TD-wzw commented 3 years ago

I seem to have found the problem. The NCNN model you provided can work normally. Is there any unknown operation during the conversion