Maxim Berman

Did you try disabling power save with sudo iw dev set power_state off? I have the same problem, I am trying to see if it helps - for now it...

I ran into this bug installing tensorboard through pip, python 2.7, linux: I get this around line 79 of the tensorboard script: ``` for mod in package_path: module_space = mod...

The easiest is to implement it as a batch sampler in PyTorch. I can share the implementation some time in the future, it will likely be in February.

That slowdown seems quite drastic, do you have e.g. many categories or images to evaluate in one step? I suspect that dedicated CUDA kernel would speed up the implementation a...

I did some profiling. It seems in tensorflow the `tf.cumsum` operation is extremely slow on GPU, and takes a huge amount of time (~99% of the total time). In Pytorch,...

This python notebook summarizes the problem: [tensorflow/profile_ops.ipynb](https://github.com/bermanmaxim/LovaszSoftmax/blob/ae50c067d5b5b701ed619b5f536b9844c1bfa0b2/tensorflow/profile_ops.ipynb) The `cumsum` operation is ~4000x slower in tensorflow vs pytorch for typical number of pixels/batch. https://github.com/tensorflow/tensorflow/issues/813 mention that current implementations of cumsum in...

After looking more into it, it seems the easiest way is to create a custom tensorflow op using [cub exclusive sum](https://nvlabs.github.io/cub/structcub_1_1_device_scan.html#a02b2d2e98f89f80813460f6a6ea1692b) instead of the native `tf.Cumsum` operation. Note that there...

Thanks for the pointer @ekelsen. Closing this issue

I don't have time to investivate into tensorflow issues for now but I am at least reopening the issue.

Hi @sourabhyadav, we will provide the trained weights later on. Keep in mind though that it won't be in caffe, I've been using my "ENet pytorch" implementation as described in...