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
5.75k stars 1.04k forks source link

caffe model of nanodet #135

Open Kyrie-Chen opened 3 years ago

Kyrie-Chen commented 3 years ago

Does anyone can help to provide the NanoDet caffe model (especially the nanodet_m.prototxt) ?

RangiLyu commented 3 years ago

Sorry, the Interp Layers in PAN are not supported in Caffe. You need to modify the model such as replace the backbone and neck if you want to run nanodet with Caffe.

Kyrie-Chen commented 3 years ago

Sorry, the Interp Layers in PAN are not supported in Caffe. You need to modify the model such as replace the backbone and neck if you want to run nanodet with Caffe.

I have modified the model and get caffe model, and convert to ncnn. But I don't know how to set the parameters of post-process in cafffe. I have seen the ncnn model you provided, and I want to know how to set the parameters of post-process in caffe which shows in the following image. Could you please provide it?

post-process in ncnn

dygcjlu commented 3 years ago

Sorry, the Interp Layers in PAN are not supported in Caffe. You need to modify the model such as replace the backbone and neck if you want to run nanodet with Caffe.

I have modified the model and get caffe model, and convert to ncnn. But I don't know how to set the parameters of post-process in cafffe. I have seen the ncnn model you provided, and I want to know how to set the parameters of post-process in caffe which shows in the following image. Could you please provide it?

post-process in ncnn

Hi Kyrie-Chen, could you kindly show me how to modify the model? I am trying to convert the model to caffe format, but failed.

Kyrie-Chen commented 3 years ago

Sorry, the Interp Layers in PAN are not supported in Caffe. You need to modify the model such as replace the backbone and neck if you want to run nanodet with Caffe.

I have modified the model and get caffe model, and convert to ncnn. But I don't know how to set the parameters of post-process in cafffe. I have seen the ncnn model you provided, and I want to know how to set the parameters of post-process in caffe which shows in the following image. Could you please provide it? post-process in ncnn

Hi Kyrie-Chen, could you kindly show me how to modify the model? I am trying to convert the model to caffe format, but failed.

Sorry for the late reply. If you want to convert the model to caffe format, you should modify the upsample layer to deconvolution and convolution layer in the "fpn.py". In addition, you should remove the split layer in the "NanoDetHead.py".