C-3-Framework
C-3-Framework copied to clipboard
gjy3035
Dimension Error! SHHA!
Hi, thanks for your excellent works and discussion.
When I run the train.py for SHHA datasets, I got a dimension error(VGG net). But only happened when I run for SHHA datasets.
RuntimeError: The size of tensor a (376) must match the size of tensor b (377) at non-singleton dimension 1
Can you please give me a suggestion？
I think the preprocessing has some problems. Because some images still have the odd length or width. I find there are some images's original shape like [1024, 713]. These images may not be resize.
Use https://github.com/gjy3035/C-3-Framework/blob/python3.x/datasets/QNRF/preapre_QNRF.m 投process SHHA.
I have this same problem and I have been tried to run prepare_QNRF, but get an error which don't have annpoints function . Can u help me or provide prepared datasets
@muzishen https://mailnwpueducn-my.sharepoint.com/:f:/g/personal/gjy3035_mail_nwpu_edu_cn/EkxvOVJBVuxPsu75YfYhv9UBKRFNP7WgLdxXFMSeHGhXjQ?e=IdyAzA
@gjy3035 Hi ! Thanks for your timely reply. I tried the data set you provided and ran your project, but the same problem occurred. Has the SHHA been updated? I wonder if the SHHA you provided still has an odd scale
You can find the annpoints in preapre_SHHB.m. annPoints = image_info{1}.location; You can try to modify it.
你好，我使用prepare_SHHB.m文件对SHHA中的图像进行resize，发现并没有任何改变，仔细看应该的.m中的程序有问题。所以我直接把每个图像的size变成standard_size = [768,1024] ,然后density对应相乘长、宽前后的比值，请问这种做法会影响groundtruth准确性吗？
@ydlu 上海B自身尺寸就是768x1024，因此不会有任何变化. 量化后准确性基本不会变化。 @muzishen We will reproduce the generation of the density map from the dot map or points using conv layer. So we have no plan to provide the processed density data.
@ydlu 抱歉，看错了你的评论。使用prepare_QNRF.py去resize即可。参见：https://github.com/gjy3035/C-3-Framework/issues/54#issuecomment-563015061
which net do you use? find the net.py you use in ./models,:add shape = x.shape at the beginning of forward func and use: x = F.interpolate(x, size=[shape[2], shape[3]]) instead of : x = F.upsample(x,scale_factor=8)
I fixed this error with this solution.hope it will be useful for you.
Cheers!
Hello.Have you obtained the correct data set? Can you send me a copy? I have never used Matlab.. Thank you.
You can find the annpoints in preapre_SHHB.m. annPoints = image_info{1}.location; You can try to modify it.
Have you obtained the correct data set? Can you send me a copy? I have never used Matlab.. Thank you
@gjy3035 Hi ! Thanks for your timely reply. I tried the data set you provided and ran your project, but the same problem occurred. Has the SHHA been updated? I wonder if the SHHA you provided still has an odd scale
Have you obtained the correct data set? Can you send me a copy? I have never used Matlab.. Thank you
Hi, thanks for your excellent works and discussion.
When I run the train.py for SHHA datasets, I got a dimension error(VGG net). But only happened when I run for SHHA datasets.
RuntimeError: The size of tensor a (376) must match the size of tensor b (377) at non-singleton dimension 1
Can you please give me a suggestion？
Have you obtained the correct data set? Can you send me a copy? I have never used Matlab.. Thank you
Why do I still get an error when I directly use the SHHA dataset file in csv format provided by the author? RuntimeError: The size of tensor a (672) must match the size of tensor b (683) at non-singleton dimension 0
