Biao Zhang
Biao Zhang
> @1zb , Thanks for sharing the preprocessed data and releasing the code. It seems that google drive has limited the download yet, and currently we can not download the...
Hi, @ran337287. You can say there is only one kernel with shape `(4, 3, kH, kW)`, or four kernels with shape `(1, 3, kH, kW)`. Both are acceptable. So yes,...
It depends on the dataset you use. For ShapeNet, I do not have the statistics now. I will reach you later.
In [original paper](https://arxiv.org/abs/1703.06211), they use deform conv in res5 which has spatial size of 7x7. So here I think you should use it after several strided convolution layers or spatial...
Backward is very slow. Forward is okay. Table 4 from the paper gives the forward time.
Thanks for the feedback. I don't have time to test it right now. Can you provide the `gradcheck` code snippets you are using?
Hi @WillSuen. I have no solution to avoid this problem right now. But I use this module in my project and it works as expected.
Did you skip the compiling step? ``` sh make.sh CC=g++ python build.py ```
这个测试用到的显存不超过500M，你得检查下显存占用
No. I don't have enough computational power to do this.