Joshua Z. Zhang

Ok, the line you pointed actually does nothing but instantiate an instance of videoLoader, basically only some header and proprocessing is done. You need to measure the real time elapsed...

@jacobssy I don't think cmake is supposed to build libnvcuvid.so, as it's shipped with CUDA toolkits. Can you post the version of cmake and the cuda/nvidia driver versions?

I don't think you will benefit from gpu decoding if you use `num_worker>0` for data loading because it's using multiple cpu core to speed up the loading which will significantly...

@DelightRun Are you using gpu for training or inference? If so, data need to be copied to gpu again from batch input.

do you happen to use the nvidia-docker and what's the version?

conflicting c libs might cause the error, since deadlocks are too difficult to track I can only imagine something goes wrong in python call.

Can you print the keyframes of the testing video you have? You can call `vr.get_key_indices()`. Another potential cause is the new RTX 2000 series with tensor cores. I haven't really...

> Okay the problem was introduced here: [7b6c0e9](https://github.com/dmlc/decord/commit/7b6c0e9a4d404087fc1fcede58b3720bb67d425b) > > [584e5f8](https://github.com/dmlc/decord/commit/584e5f811d0d50295268ecdf804873fa2bfc685a) > > ``` > 3645 frames, elapsed time for sequential read: 2.1434860229492188 > 300 frames, elapsed time for random...

@frankier and also my own test using the same video you have provided shown a very different result: Results on mac with (Intel(R) Core(TM) i7-7660U CPU @ 2.50GHz) master/HEAD ```...

Anyway I am reverting the frame based seeking to ts based in #115, I will keep this thread open as I haven't found the root cause of the issue and...