Closed only-zzx closed 3 years ago
Yes, these models are not suitable for training with random=1, because it will require a lot of GPU-memory, and decreasing mini-batch size will reduce accuracy a lot.
Yes, these models are not suitable for training with random=1, because it will require a lot of GPU-memory, and decreasing mini-batch size will reduce accuracy a lot.
ok,thanks for your answer
I reand how to improve object detection and change random=1 in my .cfg-file then the error :Assertion "0"failed happend, i think this error about the networksize,during upsample anyone can help me about that?