I have custom trained weights/config files for yolov3 tiny with relu. This works fine with darknet directly. However when I pass the same weights and config file to Detector colab crashes and restarts with no error message.
I can run your colab example fine.
I have exactly the same issue with pydarknet. It seems to be related to number of classes. It runs with 80 classes though returns nonsense. Reducing this to 1 which was used in training and it crashes.