caffe2 converted model performance

[juvebogdan]

Hello,

I trained a model that works fine when I run it in pytorch. But caffe2 converted model makes mistakes on some example images. It just misses detecting some categories. When I run the same example in jupyter notebook using predictor all works fine.

How can I fix this?