mitmul / pynvvl Goto Github PK

View Code? Open in Web Editor NEW

101.0 10.0 12.0 1.33 MB

A Python wrapper of NVIDIA Video Loader (NVVL) with CuPy for fast video loading with Python

License: MIT License

Shell 10.76% Python 89.24%

nvvl cupy video-processing gpu nvidia-video-loader video python numpy cuda

pynvvl's People

Contributors

Stargazers

Watchers

Forkers

g-votte rezoo sundoge pingchuanma rcalland mannykayy bityangke xuhuaze707313 firedfree anyvisionltd lemonyhw ajunlonglive

pynvvl's Issues

Cuda 9.2 support

Hi. Is there any plan to support cuda 9.2 soon ? Thx

multi-gpu or distributed training failed

Hi,

I try to use pynvvl to do multi-GPU training.

But it seems that it will fail.

CUDA runtime error 3 at line 15 in file /nvvl/src/detail/Decoder.cpp: initialization error CUDA runtime error 3 at line 21 in file /nvvl/src/detail/Decoder.cpp: initialization error
Could you give me some hints?

How can I transfer the returned cupy data to pytorch Tensor data for model input

torch.utils.data.Dataloder with num_workers=4, get stuck, why?

here is the code
dt = VideoIter(config.NTU, config.test_set, sampler=None, gpu_id=7)
dl = data.DataLoader(dt, 128, num_workers=4)

but, when num_workers=0, this code woks fine.

frame step support, get video clip with frame jump steps

layer.desc.stride.x = 1 layer.desc.stride.y = width * layer.desc.stride.x layer.desc.stride.c = layer.desc.stride.y * height layer.desc.stride.n = layer.desc.stride.c * channels*frame_stride
Deep Learning in videos often choose frames with a certain steps,namely jump frames. so I think it's better to support a frame stride parameters

IndexError: vector::_M_range_check

When using pynvvl, the following error occurs repeatedly, although the exception is ignored and the execution can continue successfully.
I guess it's more like CuPy error, so please close this issue if it's inappropriate to report here.

Thanks!

Exception ignored in: 'cupy.cuda.memory.PooledMemory.__dealloc__'
Traceback (most recent call last):
  File "cupy/cuda/memory.pyx", line 544, in cupy.cuda.memory.PooledMemory.free
  File "cupy/cuda/memory.pyx", line 816, in cupy.cuda.memory.SingleDeviceMemoryPool.free
  File "cupy/cuda/memory.pyx", line 830, in cupy.cuda.memory.SingleDeviceMemoryPool.free
  File "cupy/cuda/memory.pyx", line 679, in cupy.cuda.memory.SingleDeviceMemoryPool._remove_from_free_list
IndexError: vector::_M_range_check
Traceback (most recent call last):
  File "cupy/cuda/memory.pyx", line 544, in cupy.cuda.memory.PooledMemory.free
  File "cupy/cuda/memory.pyx", line 816, in cupy.cuda.memory.SingleDeviceMemoryPool.free
  File "cupy/cuda/memory.pyx", line 835, in cupy.cuda.memory.SingleDeviceMemoryPool.free
  File "cupy/cuda/memory.pyx", line 679, in cupy.cuda.memory.SingleDeviceMemoryPool._remove_from_free_list
IndexError: vector::_M_range_check
Exception ignored in: 'cupy.cuda.memory.PooledMemory.__dealloc__'
Traceback (most recent call last):
  File "cupy/cuda/memory.pyx", line 544, in cupy.cuda.memory.PooledMemory.free
  File "cupy/cuda/memory.pyx", line 816, in cupy.cuda.memory.SingleDeviceMemoryPool.free
  File "cupy/cuda/memory.pyx", line 835, in cupy.cuda.memory.SingleDeviceMemoryPool.free
  File "cupy/cuda/memory.pyx", line 679, in cupy.cuda.memory.SingleDeviceMemoryPool._remove_from_free_list
IndexError: vector::_M_range_check
Traceback (most recent call last):
  File "cupy/cuda/memory.pyx", line 544, in cupy.cuda.memory.PooledMemory.free
  File "cupy/cuda/memory.pyx", line 816, in cupy.cuda.memory.SingleDeviceMemoryPool.free
  File "cupy/cuda/memory.pyx", line 835, in cupy.cuda.memory.SingleDeviceMemoryPool.free
  File "cupy/cuda/memory.pyx", line 679, in cupy.cuda.memory.SingleDeviceMemoryPool._remove_from_free_list
IndexError: vector::_M_range_check
Exception ignored in: 'cupy.cuda.memory.PooledMemory.__dealloc__'
Traceback (most recent call last):
  File "cupy/cuda/memory.pyx", line 544, in cupy.cuda.memory.PooledMemory.free
  File "cupy/cuda/memory.pyx", line 816, in cupy.cuda.memory.SingleDeviceMemoryPool.free
  File "cupy/cuda/memory.pyx", line 835, in cupy.cuda.memory.SingleDeviceMemoryPool.free
  File "cupy/cuda/memory.pyx", line 679, in cupy.cuda.memory.SingleDeviceMemoryPool._remove_from_free_list
IndexError: vector::_M_range_check

(loooooong errors)

hi,how to release the gpu memory after call the function 'NVVLVideoLoader'?

loader = pynvvl.NVVLVideoLoader(device_id=0, log_level='error')
video = loader.read_sequence(video_root).get()

With my dataloader code, the gpu memory is increasing progressively.
what should i do to release gpu memory?

version GLIBCXX_3.4.21 not defined in file libstdc++.so.6 with link time reference

Hi,

Thanks for providing this python wrapper for nvvl.

When I am trying to import pynvvl in python, I am facing the following error:

File "", line 1, in
File "pynvvl/init.py", line 1, in
from _nvvl import NVVLVideoLoader # NOQA
ImportError: /venv_v2/lib/python2.7/site-packages/pynvvl/_lib/libnvvl.so: symbol _ZNSt7__cxx1112basic_stringIcSt11char_traitsIcESaIcEE9_M_appendEPKcm, version GLIBCXX_3.4.21 not defined in file libstdc++.so.6 with link time reference

Any thoughts about this issue?

Thank you

Pynvvl or nvvl PyTorch wrapper?

The readme doesn't explain at all what's the motivation for using this wrapper instead of the PyTorch wrapper provided in nvvl. I myself don't know whether this wrapper would better fit my needs.

Sometime got stuck

I used the following snippet, but sometimes got stuck at some videos.
I checked the Memory and Graphics Card, neither of them is full.

	num=100000
	start=time.time()
	video_list=sorted(os.listdir(path))[:num]
	num = min(len(video_list),num)
	for video_name in video_list:
		video_path = os.path.join(path,video_name)
		print (video_path)
		loader = pynvvl.NVVLVideoLoader(device_id=0, log_level='error')
		n_frames = loader.frame_count(video_path)
		print('Number of frames:', n_frames)
	duration = time.time()-start
	print(duration,duration/num)

Scale Height and Scale Width Don't Scale Video Frames

I am using nvidia's cuda toolkit 9.2 with cudnn 7 on Ubuntu 18.04, and when I load videos in, everything works fine, except when trying to use the scale_height and scale_width arguments. When I use these, the videos are still the same shape as before.

CUDA 10.0 pynvvl support?

Is cuda 10.0 on the roadmap?

pynvvl-cuda90 (0.0.3a2) - PyNVVL: A Python wrapper for NVIDIA Video Loader (NVVL) with CuPy
pynvvl-cuda80 (0.0.3a2) - PyNVVL: A Python wrapper for NVIDIA Video Loader (NVVL) with CuPy
pynvvl-cuda92 (0.0.3a2) - PyNVVL: A Python wrapper for NVIDIA Video Loader (NVVL) with CuPy
pynvvl-cuda91 (0.0.3a2) - PyNVVL: A Python wrapper for NVIDIA Video Loader (NVVL) with CuPy

mitmul / pynvvl Goto Github PK

pynvvl's People

Contributors

Stargazers

Watchers

Forkers

pynvvl's Issues

Recommend Projects

Recommend Topics

Recommend Org

Jobs