Jorge Orfanidis
Jorge Orfanidis
What is the score (accuracy, F1 etc) you are getting? And how much has it improved by the various methods?
From my experience the low loss on training set does not guarantee a correspondent low mAP in test set. Maybe you got higher mAP on test set at a higher...
Well, as far as I decipher: The first digit is always 1 and it denotes that this frame is present (and not absent that is). Not really helpful indead. The...
You need to provide a path to the an actual video in your system since "test.mp4" used by the author to test the model in not provided. Lines to be...
I think SSD typically has higher loss value than Faster RCNN for the same dataset from my experience. I am not sure this can be ported directly to better performance....
I would guess that he certainly used pretrained model (it's the typical approach after all). I am not sure the performance gap is reasonable between training from scratch and using...