usr/local/lib/python3.5/dist-packages/torch/include/torch/csrc/api/include/torch/optim/sgd.h:48:17: required from here usr/local/lib/python3.5/dist-packages/torch/include/torch/csrc/api/include/torch/nn/options/pooling.h:428:8: required from 'void torch::nn::Cloneable::clone_(torch::nn::Module&, const c10::optional&) ' usr/local/lib/python3.5/dist-packages/torch/include/c10/util/Optional.h: In instantiation of 'c10::optional& c10::optional::operator=(c10::optional&) ': Thanks, can you please make a fork of vid2vid / flownet2 that does these things automatically or a Docker image? I really appreciate that. It should now compile successfully with PyTorch 1.0.0. * Rebuild flownet2 after making those changes with `bash install.sh` inside the `models/flownet2_pytorch` folder. * You don't need to do all of them, only really adding `#include ` and then replacing `at::globalContext().getCurrentCUDAStream()` with `at::cuda::getCurrentCUDAStream()` in each of the three sub-packages - `channelnorm-cuda`, `correlation-cuda`, and `resample2d-cuda`. * Go to `models/flownet2_pytorch` and make the changes in this pull request: You'll get some compilation errors and warnings as the current version of flownet2 is still designed for the earlier version of Torch. * Then, run `python scripts/download_flownet2.py`. * Modify the `download_flownet2.py` file from `if torch._version_ = '0.4.1':` to `if torch._version_ = '1.0.0':` * `git clone` the () repository directly into the `models` folder in vid2vid to `flownet2_pytorch` (recommended)
* Continue to download the datasets with `python scripts/download_datasets.py`.
* First, download all of the dependencies needed for vid2vid, substituting CUDA 10 and PyTorch 1.0.0 instead of CUDA 9.x and PyTorch 0.4.x. Thanks for the original PR to flownet2 - saved me a lot of time from trying to dig through the new PyTorch API changes there. If there's any interest, I can make a fork of vid2vid / flownet2 that does these things automatically as well or a Docker image. I was able to get this working on CUDA 10, PyTorch 1.0.0, on a RTX 2080 TI. You can now run vid2vid following the rest of the instructions. Rebuild flownet2 after making those changes with bash install.sh inside the models/flownet2_pytorch folder. You don't need to do all of them, only really adding #include and then replacing at::globalContext().getCurrentCUDAStream() with at::cuda::getCurrentCUDAStream() in each of the three sub-packages - channelnorm-cuda, correlation-cuda, and resample2d-cuda.Go to models/flownet2_pytorch and make the changes in this pull request:
I thought this might help a few people, though feel free to remove this if you think it doesn't belong in the issues tracker.