Some parameters in the checkpoint have a different shape

[andreazuna89]

Dear authors,
thanks for the great work. I am trying your eval code using pre-trained weights (for example yolact_edge_resnet50_54_800000.pth) but I get a warning in the beginning which seems causing no output predictions in the image:

Some parameters in the checkpoint have a different shape in the current model, and are initialized as they should be: prediction_layers.0.conf_layer.weight, prediction_layers.0.conf_layer.bias, semantic_seg_conv.weight, semantic_seg_conv.bias

The same error is given using all the available pre-trained weights. The same code was running well before you have added the deform conv feature to the code. Do you think you have missed something in the extension of the code?

Thanks a lot

[upupdayday]

Met the same warning when inferencing on cpu

[Hogushake]

I also had a similar problem, but the problem was that the config was not recognized properly.

If you run it from the prompt, just enter the correct --config.