Zijian Zhou
@leejiajun I changed my dataset,but I can train it on the original net at first,but when I changed the net, it can not work.The same dataset, but can not work...
@twinsyssy1018 In YOLOv2, we got the anchor by K-Means, which was explained by the paper:[https://arxiv.org/abs/1612.08242](url). You can use this to generate anchors fit your own dataset.
@hzh8311 Have you solved this problem yet? I encountered this problem too. @YuwenXiong Is there any method except using smaller learning rate?
@mursalal @hzh8311 I have solved this problem. I think you shoud check your training data again and again, write some code to avoid some extreme values. Besides, you can check...
@YuwenXiong Looking forward to your reply.
‘len(roidb)’ means the number of instance. '4.83' means how many epoches to reduce lr.
Did you do the same experiment on MXNet? Will there be such a phenomenon? @louyanyang
@louyanyang Could you show some log?
It looks like RCNN is adjusting to something, Train more epoches, it may rise. @louyanyang
``` def get_resnet_50(self): sym = mx.sym.load('model/pretrained_model/resnet-50-symbol.json') return sym.get_internals()['data'], sym.get_internals()['relu1_output'] ``` Hope this can help you.