Load pretrained model problem

[yhydhx]

Thank you for your amazing work. I trained a retinanet model on my customized dataset. The detection works well on the RetinaNet model. However, I load the pretrained model in CTracker and the detection didn't work in CTracker. I checked my code and I found that these layers are not successfully loaded in the CTracker model because of mismatched size: #1. regressionModel.conv1.weight 256,512,3,3 => 256,256,3,3 #2. regressionModel.output.weight 36,256,3,3 => 8,256,3,3 #3. regressionModel.output.bias 8 => 36 #4. classificationModel.conv1.weight 256,512,3,3 => 256,256,3,3 #5. classificationModel.output.weight 1,256,3,3=> 9,256,3,3 I tried to change the parameters in either Ctracker model or RetinaNet model. But none of them can run successfully. Could you tell me 1. how to train your RetinaNet with default setting? Should I change feature size from 256 to 512? Should I change the num_anchors from 9 to 1? Did you load all the weight in your RetinaNet ?

Thanks so much! Harold.