Error in MaskOutput.infer_type: Traceback (most recent call last): File "/disk1/multitask/mx-maskrcnn/incubator-mxnet/python/mxnet/operator.py", line 723, in infer_storage_type_backward_entry assert num_tensor == (2 * n_in + 2 * n_out + n_aux) AssertionError

Open devinxzhang opened this issue 8 years ago • 3 comments

I met this bug in the step: "######TRAIN RCNN WITH IMAGENET INIT AND RPN DETECTION" when init the network parameters. Who could solve it? THX!

Nov 18 '17 03:11 devinxzhang

I tried that: left(num_tensor)=9 while right(2 * n_in + 2 * n_out + n_aux) =10. How does this come? I modified nothing but the learning rate and gpu_num. @huazehuang @Zehaos

Nov 20 '17 02:11 devinxzhang

Try to use branch in this repo to compile and train.

Dec 01 '17 12:12 kaiyuyue

I fixed this issue with custom op in https://github.com/apache/incubator-mxnet/pull/8721. And the sparse support for custom op is reverted in https://github.com/apache/incubator-mxnet/pull/8733. So it should be ok with the latest mxnet.

Dec 07 '17 18:12 ZiyueHuang