Ali Darwish, Ph.D.

@cocoza4 awesome! yeah, you certainly have to add 1 to num_classes. I'm almost uploading the code to aws. Will keep you posed. The only problem is that in my keras...

@cocoza4 there is something wrong here in SSD_training you have # some constants ``` CLASSES = ['background' 'car' 'pedestrian'] # CLASSES = ['car' 'pedestrian'] NUM_CLASSES = 4 # 1 added...

@cocoza4 something isn't right here, I'll look it up!

@cocoza4 SO I'm testing your generator and the training script and it does not show a progress bar. I know its training from monitoring the GPU resources. Do you have...

@cocoza4 Huh! this looks pretty ugly. Is the code that produced this output updated on github? I'm working on the background issue too.

@cocoza4 Ok, I'm training the network and we should see the outcome tomorrow. Here is how I'm fitting but better look at the whole code when I uploaded "if it...

A very important thing which is driving your accuracy very low is the way the training and validation is split. The way both you and I are doing it is...

@cocoza4 will you please include accuracy in your model when you train, it's always easier to look at that.

Here https://mega.nz/#F!7RowVLCL!q3cEVRK9jyOSB9el3SssIA

@liuwei16 I tried to send you a copy but your email rejected it. It's the big size I guess. If you still need it I will upload it to a...