Vision: when is 0.5.0 expected to be released ?

[danFromTelAviv]

I would like to use the imagenet dataset.
When I try to install using setup.py I get the following error:

creating build/temp.linux-x86_64-3.6/home/dan/vision/torchvision/csrc/cpu
x86_64-linux-gnu-gcc -pthread -DNDEBUG -g -fwrapv -O2 -Wall -g -fstack-protector-strong -Wformat -Werror=format-security -Wdate-time -D_FORTIFY_SOURCE=2 -fPIC -I/home/dan/vision/torchvision/csrc -I/usr/local/lib/python3.6/dist-packages/torch/include -I/usr/local/lib/python3.6/dist-packages/torch/include/torch/csrc/api/include -I/usr/local/lib/python3.6/dist-packages/torch/include/TH -I/usr/local/lib/python3.6/dist-packages/torch/include/THC -I/usr/include/python3.6m -c /home/dan/vision/torchvision/csrc/vision.cpp -o build/temp.linux-x86_64-3.6/home/dan/vision/torchvision/csrc/vision.o -DTORCH_API_INCLUDE_EXTENSION_H -DTORCH_EXTENSION_NAME=_C -D_GLIBCXX_USE_CXX11_ABI=0 -std=c++11
In file included from /home/dan/vision/torchvision/csrc/vision.cpp:8:0:
/home/dan/vision/torchvision/csrc/ROIAlign.h:80:24: error: ‘torch::autograd::AutogradContext’ has not been declared
 using torch::autograd::AutogradContext;
                        ^~~~~~~~~~~~~~~
/home/dan/vision/torchvision/csrc/ROIAlign.h:84:58: error: expected template-name before ‘<’ token
 class ROIAlignFunction : public torch::autograd::Function<ROIAlignFunction> {
                                                          ^
/home/dan/vision/torchvision/csrc/ROIAlign.h:84:58: error: expected ‘{’ before ‘<’ token
/home/dan/vision/torchvision/csrc/ROIAlign.h:84:58: error: expected unqualified-id before ‘<’ token
/home/dan/vision/torchvision/csrc/ROIAlign.h: In function ‘at::Tensor roi_align(const at::Tensor&, const at::Tensor&, double, int64_t, int64_t, int64_t)’:
/home/dan/vision/torchvision/csrc/ROIAlign.h:140:28: error: incomplete type ‘ROIAlignFunction’ used in nested name specifier
   return ROIAlignFunction::apply(
                            ^~~~~
In file included from /home/dan/vision/torchvision/csrc/vision.cpp:9:0:
/home/dan/vision/torchvision/csrc/ROIPool.h: At global scope:
/home/dan/vision/torchvision/csrc/ROIPool.h:70:24: error: ‘torch::autograd::AutogradContext’ has not been declared
 using torch::autograd::AutogradContext;
                        ^~~~~~~~~~~~~~~
/home/dan/vision/torchvision/csrc/ROIPool.h:74:57: error: expected template-name before ‘<’ token
 class ROIPoolFunction : public torch::autograd::Function<ROIPoolFunction> {
                                                         ^
/home/dan/vision/torchvision/csrc/ROIPool.h:74:57: error: expected ‘{’ before ‘<’ token
/home/dan/vision/torchvision/csrc/ROIPool.h:74:57: error: expected unqualified-id before ‘<’ token
/home/dan/vision/torchvision/csrc/ROIPool.h: In function ‘std::tuple<at::Tensor, at::Tensor> roi_pool(const at::Tensor&, const at::Tensor&, double, int64_t, int64_t)’:
/home/dan/vision/torchvision/csrc/ROIPool.h:125:34: error: incomplete type ‘ROIPoolFunction’ used in nested name specifier
   auto result = ROIPoolFunction::apply(
                                  ^~~~~
error: command 'x86_64-linux-gnu-gcc' failed with exit status 1

It seems like 0.5.0 has been in the oven for a few months now and I would rather wait for a stable pip install if it's coming soon.

Thanks for all your hard work on behalf of the community.

[fmassa]

Hi,

ImageNet is present in both TorchVision 0.3 and 0.4, which are both available in pip / conda, and works with a stable PyTorch.

Would it be enough for you?

[danFromTelAviv]

i have 0.4.0 installed
i get the following error
"'module \'torchvision.datasets\' has no attribute \'Imagenet\''"

[pmeier]

That is because it is titled ImageNet with a capital N. See the official documentation or the implementation in v.0.4.0:

https://github.com/pytorch/vision/blob/a1ed206ff422f7e9a49c63d4b507ec3aa5fb770d/torchvision/datasets/imagenet.py#L26-L27

[danFromTelAviv]

perfect - that works! Thank you very much!

[fmassa]

FYI, the next release of torchvision should be out by the end of the month, but it might be named 0.4.1 instead of 0.5, TBD.

I'm closing the issue, but let us know if you have further questions