Skip to content

Latest commit

 

History

History
 
 

samples

README.md

rocPyDecode samples

Prerequisites

The torch python sample requires pytorch for ROCm, which can be installed as follow:

  • If using bare-metal, sudo access is needed.
    pip3 install --pre torch torchvision torchaudio --index-url https://download.pytorch.org/whl/nightly/rocm6.2
    sudo reboot
  • If using a docker environment or any system with root access, no need for reboot.
    pip3 install --pre torch torchvision torchaudio --index-url https://download.pytorch.org/whl/nightly/rocm6.2

The performance sample requires python HIP, which can be installed as follows:

python3 -m pip install --upgrade pip
python3 -m pip install -i https://test.pypi.org/simple hip-python

videodecode.py

This sample demuxes & decode frames from a video file, and optionally saves the frames to a file.
To run this python sample script, you need to provide input video file full path name, other arguments are optional.

Arguments

The following are full list of arguments that can be passed to the sample.

-h, --help                                       : Show detail help message and exit
-i INPUT, --input INPUT                          : Input File Path - required
-o OUTPUT, --output OUTPUT                       : Output File Path - optional
-d DEVICE, --device DEVICE                       : GPU device ID - optional, default - 0
-m MEM_TYPE, --mem_type MEM_TYPE                 : Memory Type of output surfce - 0: Internal 1: dev_copied 2: host_copied 3: MEM not mapped - optional, default 0
-z ZERO_LATENCY, --zero_latency ZERO_LATENCY     : Force zero latency - [options: yes,no], default: no
-crop CROP_RECT, --crop_rect CROP_RECT           : Crop rectangle (left, top, right, bottom) - optional, default: None (no cropping)
-s SEEK, --seek SEEK                             : seek this number of frames, optional, default: no seek
-sm SEEK_MODE, --seek_mode SEEK_MODE             : seek mode, 0 - by exact frame number, 1 - by previous key frame, optional, default: 1 - by previous key frame
-sc SEEK_CRITERIA, --seek_criteria SEEK_CRITERIA : seek criteria, 0 - by frame number, 1 - by time stamp, optional, default: 0 - by frame number
-resize RESIZE_DIM RESIZE_DIM, --resize_dim RESIZE_DIM RESIZE_DIM : Width & Height of new resized frame, optional, default: no resizing

videodecodeperf.py

This sample demuxes & decode frames from a video file on multiple processes. User can define the number of parallel jobs to observe performance scaling.
To run this python sample script, you need to provide input video file full path name, other arguments are optional.

Arguments

The following are full list of arguments that can be passed to the sample.

-h, --help                                                  : Show detail help message and exit
-i INPUT, --input INPUT                                     : Input File Path - required
-d DEVICE, --device DEVICE                                  : GPU device ID - optional, default - 0
-m MEM_TYPE, --mem_type MEM_TYPE                            : Memory Type of output surfce - 0: Internal 1: dev_copied 2: host_copied 3: MEM not mapped - optional, default 3
-t Number of Processes, --num_process Number of Processes   : Number of Processes - optional, default 4

videodecodergb.py

This sample demuxes & decode frames from a video file, and convert it to an rgb frame. Optionally you can save the rgb frames to a file. The output is the rgb frame with the format the user specifies with -of or as default 'rgb' if not specified. To run this python sample script, you need to provide input video file full path name, other arguments are optional.

Arguments

The following are full list of arguments that can be passed to the sample.

-h, --help                                    : Show detail help message and exit
-i INPUT, --input INPUT                       : Input File Path - required
-o OUTPUT, --output OUTPUT                    : Output File Path - optional
-d DEVICE, --device DEVICE                    : GPU device ID - optional, default - 0
-m MEM_TYPE, --mem_type MEM_TYPE              : Memory Type of output surfce - 0: Internal 1: dev_copied 2: host_copied 3: MEM not mapped - optional, default 0
-z ZERO_LATENCY, --zero_latency ZERO_LATENCY  : Force zero latency - [options: yes,no], default: no
-crop CROP_RECT, --crop_rect CROP_RECT        : Crop rectangle (left, top, right, bottom) - optional, default: None (no cropping)
-of RGB_FORMAT, --rgb_format RGB_FORMAT       : Rgb Format to use - 1:bgr, 3:rgb, converts decoded YUV frame to Tensor in RGB format, optional, default: 3

videodecodergb.py

This sample demuxes & decode frames from a video file, and convert it to an rgb frame. Optionally you can save the rgb frames to a file. The output is the rgb frame with the format the user specifies with -of or as default 'rgb' if not specified. To run this python sample script, you need to provide input video file full path name, other arguments are optional.

Arguments

The following are full list of arguments that can be passed to the sample.

-h, --help                                    : Show detail help message and exit
-i INPUT, --input INPUT                       : Input File Path - required
-o OUTPUT, --output OUTPUT                    : Output File Path - optional
-d DEVICE, --device DEVICE                    : GPU device ID - optional, default - 0
-m MEM_TYPE, --mem_type MEM_TYPE              : Memory Type of output surfce - 0: Internal 1: dev_copied 2: host_copied 3: MEM not mapped - optional, default 0
-z ZERO_LATENCY, --zero_latency ZERO_LATENCY  : Force zero latency - [options: yes,no], default: no
-crop CROP_RECT, --crop_rect CROP_RECT        : Crop rectangle (left, top, right, bottom) - optional, default: None (no cropping)
-of RGB_FORMAT, --rgb_format RGB_FORMAT       : Rgb Format to use - 1:bgr, 3:rgb, converts decoded YUV frame to Tensor in RGB format, optional, default: 3

videodecodetorch.py

This sample demuxes & decode frames from a video file, and convert it to pytorch tensor via DLPack. Optionally you can save it to a file.
To run this python sample script, you need to provide input video file full path name, other arguments are optional.

Arguments

The following are full list of arguments that can be passed to the sample.

-h, --help                                    : Show detail help message and exit
-i INPUT, --input INPUT                       : Input File Path - required
-o OUTPUT, --output OUTPUT                    : Output File Path - optional
-d DEVICE, --device DEVICE                    : GPU device ID - optional, default - 0
-m MEM_TYPE, --mem_type MEM_TYPE              : Memory Type of output surfce - 0: Internal 1: dev_copied 2: host_copied 3: MEM not mapped - optional, default 0
-z ZERO_LATENCY, --zero_latency ZERO_LATENCY  : Force zero latency - [options: yes,no], default: no
-crop CROP_RECT, --crop_rect CROP_RECT        : Crop rectangle (left, top, right, bottom) - optional, default: None (no cropping)

videodecodetorch_yuv.py

This sample demuxes & decode frames from a video file, and convert it to pytorch tensor via DLPack, each component in a separate tensor. Optionally you can save it to a file, specifying saving the chroma plane U/V with -y no, otherwise will save the luma plane.
To run this python sample script, you need to provide input video file full path name, other arguments are optional.

Arguments

The following are full list of arguments that can be passed to the sample.

-h, --help                                    : Show detail help message and exit
-i INPUT, --input INPUT                       : Input File Path - required
-o OUTPUT, --output OUTPUT                    : Output File Path - optional
-d DEVICE, --device DEVICE                    : GPU device ID - optional, default - 0
-m MEM_TYPE, --mem_type MEM_TYPE              : Memory Type of output surfce - 0: Internal 1: dev_copied 2: host_copied 3: MEM not mapped - optional, default 0
-z ZERO_LATENCY, --zero_latency ZERO_LATENCY  : Force zero latency - [options: yes,no], default: no
-crop CROP_RECT, --crop_rect CROP_RECT        : Crop rectangle (left, top, right, bottom) - optional, default: None (no cropping)

videodecodetorch_resnet50.py

This sample demuxes & decode frames from a video file, converts each frame to pytorch tensor via DLPack.
It resizes the frame to match resnet50 model, and feed it to the model inference, printing out 5 possible predictions with precision ratio, for each frame it decodes.
To run this python sample script, you need to provide input video file full path name, other argument is optional.

Arguments

The following are full list of arguments that can be passed to the sample.

-h, --help                                    : Show detail help message and exit
-i INPUT, --input INPUT                       : Input File Path - required
-d DEVICE, --device DEVICE                    : GPU device ID - optional, default - 0