zlai0 / corrflow Goto Github PK

The .csv file for the OxUvA dataset is not present in the datas folder with the kinetics.csv folder.

Did anybody successfully installed the Pytorch Correlation module on Windows?
I get the error

  1 error detected in the compilation of "C:/Users/root/AppData/Local/Temp/tmpxft_0000bb6c_00000000-10_correlation_cuda_kernel.cpp1.ii".
  correlation_cuda_kernel.cu
  error: command 'C:\\Program Files\\NVIDIA GPU Computing Toolkit\\CUDA\\v10.0\\bin\\nvcc.exe' failed with exit status 1

caused by

C:/Logiciels/Anaconda3/envs/torch/lib/site-packages/torch/include\torch/csrc/jit/argument_spec.h(161): error: member "torch::jit::ArgumentSpecCreator::DEPTH_LIMIT" may not be initialized

I found the same issue here pytorch/extension-cpp#37, but it is a month old with no answer...

tar -xvf kinetics.tar tar: This does not look like a tar archive tar: Skipping to next header tar: Exiting with failure status due to previous errors
The same is true for the oxuva.tar file also.

Hello. I really liked your paper but I am a bit confused about how you do video segmentation. Do you encode a sequence of RGB images with the segmentation mask already applied to it? And then, you predict the next sequence of RGB images with the segmentation masks applied to them?

In the readme, we can find this:

Then you can test the output with the official Python evaluation code.

python evaluation_method.py --task semi-supervised --results_path log-path

What is evaluation_method.py and where can I find it? Let's say I want to visualize the output on the OxUvA dataset.

Are the training hyper-parameters same for OxUvA as well? 1 million iterations, with a batch size of 8? If not, what values do you use for 'epochs' (not iterations), 'lr' and 'bsize' in main_oxuva.py?

Hi, I tried to reproduce your experients on OxUvA dataset. I found the default batch size is 6 which is too large for a single 2080ti. Therefore, I used 2 GPUs and the J&F-Mean result was 40.5237, lower than the paper result(50.3).
Is it related to the batch size? Could you pls tell me what's your training batchsize on each GPU and how many GPUs you used.

Good work! I have a question about training time. How long to run the training code on OxUaV dataset with 30 epochs?

How was the ablation study without restricted attention conducted? What was used for the calculating the correlation?

Wondering how to use keypoint tracking?

Hi,

When will the full code be released?

Is the kinetics.tar file actually a tar file? It doesn't open with tar -xvf kinetics.tar, nor by assuming it's a .tar.gz and trying similar.

The OxUvA dataset also loads centroids from centroids_16k_kinetics_10000samples.npy for quantizing the color space. Shouldn't the code compute new centroids for a different dataset? Can we use the same centroids for a custom dataset that we plan to use with the code?