Error when run test_fcn32_vgg.py

[Monsterhoho]

Hello Marvin,

I have some trouble when I run your codes.

1) When I run the test_fcn32_vgg.py, it reminds me that there is no enough memory to allocate the tensor. (with gpu) However, I can run the test_fcn16_vgg.py(with gpu), it works well. And then I tried running with only cpu, both test_fcn32_vgg.py and test_fcn16_vgg.py works well.

I read the codes and it seems that test_fcn16_vgg.py may need more memory?

Is it just for that my memory of GPu is not enough (it is about 2GB) or something else?

2) I run the test code, but the result is not so good.

And you can see the part of the dog is missing, which is not like your result on github.

Thanks for your time.

Sincerely, Shuhong

[MarvinTeichmann]

Hi Shuhong,

The network is not trained on Segmentation data. So do not expect the results to be good. For trained Networks check out my KittiSeg repository. If you want to do semantic segmentation, 2GB is not enough. With 6 and more GB the model is running fine.

[MarvinTeichmann]

and the fcn32 memory needs more memory then fcn16, as it a larger upsampling filter.

[Monsterhoho]

I do not really understand it. The upsampling fileter is not very big. For fcn32, the filter from the get_decov is [64,64,20,20] For fcn16, the upsampling filter is [4,4,20,20] and [32,32,20,20]. All of them seem not much large and will not take much memory.

[MarvinTeichmann]

Computing the inference takes much more memory then just the weight size. Try other deep learning code. If you work with high resolution images, you will not get far with 2GB.