rog

@unsky Hi, thanks for your work. I have tried training ResNet50 deformable Fast RCNN end2end model on WIDER FACE DataSet, the loss seemed normally converge, but I got nothing when...

@unsky I don't think I have many bugs in my test, here is my results using your res50 faster rcnn config but replace deformable convolution by normal convolution, trained on...

@lawpdas yes, my results seems normal now on both resnet50 and VGG16

@zhanglonghao1992 You may check DCNV2 here https://github.com/msracver/Deformable-ConvNets, They have already solved a possible issue when the sampling location is outside of image boundary

@suguliwei sorry, I will not, but I can share some of my training details here, my solver.proto is like below: lr_policy: "multistep" stepvalue: 300000 stepvalue: 600000 stepvalue: 900000 base_lr: 0.045...

@chunniunai220ml 1.why multistep policy? just personal experience. 2.how to understand the stepvalue? ImageNet Classfication training set has about 1280k images，300k iteration (128 batchsize) ≈ 30 epochs 3.Inception augmentation means ImageNet...