Do you have any information on the inference speeds on CPU/GPU for the different models? Using the test script you provided and using the MobileNetV2 network as backbone for DOTA on the CPU I got 1 image prediction in a minute, does this make sense?
DOTA contains large-resolution images. In the testing phase, the images need to be cropped and tested, and then all prediction results are merged. A image in DOTA may be cropped into more than a dozen small images.
Hello, it's a great project you created.
Do you have any information on the inference speeds on CPU/GPU for the different models? Using the test script you provided and using the MobileNetV2 network as backbone for DOTA on the CPU I got 1 image prediction in a minute, does this make sense?