search

RangiLyu / nanodet

NanoDet-Plus⚡Super fast and lightweight anchor-free object detection model. 🔥Only 980 KB(int8) / 1.8MB (fp16) and run 97FPS on cellphone🔥
Apache License 2.0
5.75k stars 1.04k forks source link

Backbone shufflenetv2_0.5x RuntimeError #117

Open qiaoshh opened 3 years ago

qiaoshh commented 3 years ago

RuntimeError: Given groups=1, weight of size [96, 116, 1, 1], expected input[32, 48, 76, 76] to have 116 channels, but got 48 channels instead.

shaoshengsong commented 3 years ago

in_channels需要更改，这样就不会有错误了。https://github.com/shaoshengsong/quarkdet 具体地址是 https://github.com/shaoshengsong/quarkdet/blob/main/config/shufflenetv2_0.5x.yml

neck: name: PAN in_channels: [48, 96, 192]

qiaoshh commented 3 years ago

in_channels需要更改，这样就不会有错误了。https://github.com/shaoshengsong/quarkdet 具体地址是 https://github.com/shaoshengsong/quarkdet/blob/main/config/shufflenetv2_0.5x.yml

neck: name: PAN in_channels: [48, 96, 192]

报错，RuntimeError: Given groups=1, weight of size [96, 92, 1, 1], expected input[64, 96, 38, 38] to have 92 channels, but got 96 channels instead. 这里的in_channels怎么计算得到的啊

qiaoshh commented 3 years ago

in_channels需要更改，这样就不会有错误了。https://github.com/shaoshengsong/quarkdet 具体地址是 https://github.com/shaoshengsong/quarkdet/blob/main/config/shufflenetv2_0.5x.yml neck: name: PAN in_channels: [48, 96, 192]

报错，RuntimeError: Given groups=1, weight of size [96, 92, 1, 1], expected input[64, 96, 38, 38] to have 92 channels, but got 96 channels instead. 这里的in_channels怎么计算得到的啊

shufflenetv2_0.5x，in_channels：[48，96，192]运行正常了，谢谢！