ranchentx / efgrnet Goto Github PK

hi,
Thank you for your contribution.
I have some marginal questions. In evaluation file, what is the backbone time? and do you have any method to reduce the time ? because when i run evaluation file and found that the backbone time is maybe long.

I have read your paper,but the output objectness module are C1x and B1x,

why and how???

我用您给的代码自己训练得到的精度并没有论文中那么高，请问怎么回事。

Can I ask you some question?I want to know whether your train_coco.py contain the function of training VOC Dataset or not.

In the paper,for MS COCO dataset, the learning rate is change at 90, 120 and 140 epoch, respectively.
But in the yaml, the learning rate is change at 90, 110 ,130,150 and 160 epoch ，and added a learning rate of 0.002.

Why the learning rate is different from the paper

hello, I can't find the code of the multi-scale inference in the project, can you help me?

Hello, can you share the code of feature map visualization?

What is the solution to this error? Thanks.

RuntimeError: max_pool2d_with_indices_out_cuda_frame failed with error code 0
File "/data_1/software/yes/envs/pytorch/lib/python3.6/site-packages/torch/nn/functional.py", line 488, in _max_pool2d
input, kernel_size, stride, padding, dilation, ceil_mode)
RuntimeError: max_pool2d_with_indices_out_cuda_frame failed with error code 0
请问，这是什么原因导致呢？测试图像需要提前resize到320*320尺寸才能用吗？