Li Zhu

From the code, I found that the 75.02 mAP was obtained by multi scale training and test(scale[0.5, 1.0]), Is that true? how about the result of single scale training？

The speed of train_maskrcnn on coco

3

I use 4 GTX 1080(single image per GPU) to alternatively train mask r-cnn on COCO. When training RPN, the speed can reach 8 sample/sec. But when training mask r-cnn, it...

I modified the Initialization by:  And set the rpn_lr=0.002 with noing else changed. When using single GPU card, the speed of convergence is very fast. The RPNAcc can go...