linux installation errors

[MHRosenberg]

Hi there, your package looks very cool. Unfortunately, I was not able to follow your instructions to install it. Could it be because I was trying to install it inside a conda environment?

My system:
uname -a:
Linux SERVER_ADDRESS 5.14.0-427.35.1.el9_4.x86_64 #1 SMP PREEMPT_DYNAMIC Thu Sep 12 18:24:53 UTC 2024 x86_64 x86_64 x86_64 GNU/Linux

I've attached the terminal output below.

Thanks!

250107_tmux_output-ffmpegcvInstallFail.txt

[chenxinfeng4]

You've install the ffmpegcv by

$ pip install ffmpegcv

The ffmpegcv[cuda] is only for advanced usage. You can skip it.

pip install ffmpegcv[cuda]

[chenxinfeng4]

You can test the ffmpegcv by

# python
import ffmpegcv
vid = ffmpegcv.VideoCapture('PATH_TO_VIDEO')
ret, frame = vid.read()

[MHRosenberg]

Hmmm are you saying that pip install ffmpegcv[cuda] isn't necessary to use the GPU acceleration?

[chenxinfeng4]

not all GPU acceleration tricks need ffmpegcv[cuda].

see the TOCUDA section of the readme.

[MHRosenberg]

I'm mostly interested in using your code to crop videos. At some point I might want to read the frames with ffmpegcv, annotate them with matplotlib/opencv, and then write them to a new video. I read the TOCUDA section, but it's not clear to me if that would require the ffmpegcv[cuda] or not.

1. Would it be possible to comment on this?

2. I'd also like some installation tips (if not instructions) for the ffmpegcv[cuda] install regardless of your answer to 1.

3. (ffmpegcv follows opencv and defines the top left of the image as (0, 0), right?)

[chenxinfeng4]

The GPU can be used to

1. Read/write the video file. Functions are ffmpegcv.VideoCaptureNV, ffmpegcv.VideoWriterNV. pip install ffmpegcv
2. Crop the video region. Top left of the image is (0, 0) . ffmpegcv.VideoCapture(..., crop_xywh=xxx). pip install ffmpegcv
3. Resize the video region. ffmpegcv.VideoCapture(..., resize=xxx). pip install ffmpegcv.
4. Shortcut of image to the NVIDIA pytorch tensor, useful for deeplearning processing. ffmpegcv.toCUDA(...). pip install ffmpegcv[cuda]

The 1/2/3 programs are using the NVIDIA_DECODER and the NVIDIA_ENCODER units to accelerate processing. Which need minimal installation requirement (the NVIDA driver and a proper ffmpeg release).
The 4nd program is using the NVIDIA_CUDA unit, which need many CUDA installation requirement (CUDA, NVCC).

[MHRosenberg]

Thank you! I just noticed the image origin in your doc, my apologies.

This might also be in your docs but could you comment briefly on when to use FFmpegWriter vs FFmpegWriterNV? Your docs and code seems to suggest that FFmpegWriter does in fact use the GPU under some circumstances?

[chenxinfeng4]

The VideoWriter using CPU to encode video. While the VideoWriterNV can use GPU.
For some reason, the VideoWriterNV won't be much faster, but can save CPU usage.