gstreamer-python

Purpose

• abstraction over PyGOBJECT API for Gstreamer
• work with Gstreamer metadata
• common tools for Gstreamer pipeline management
• easy gst-python installation

Install

FOR ANY VIRTUAL ENVIRONMENT USERS, MAKE SURE YOU COMPILED PYTHON WITH --enable-shared(ESPECTIALLY FOR PYENV USERS) OR GST-PYTHON WON'T BE PROPERLY INSTALLED.

Install OS packages

• How to install Gstreamer on Ubuntu
• How to install Gstreamer Python Bindings

in-place

You need to install meson and ninja first. Usually just search meson and install it using your package manager and ninja will be automatically installed.

python3 -m venv venv

source venv/bin/activate
pip install --upgrade wheel pip setuptools
pip install --upgrade --requirement requirements.txt

./build-3rd-party.sh
./build-gst-python.sh

pip-package

You need to install meson and ninja first. Usually just search meson and install it using your package manager and ninja will be automatically installed.

Note this package will not raise an exception for any error that happends. In the case of any error occurs, you might have a broken installation(e.g. gst-python not properly installed).

pip install git+https://github.com/jackersson/gstreamer-python.git@{tag_name}#egg=gstreamer-python

### to skip ./build-gst-python.sh
pip install . -v --install-option "build_py" --install-option "--skip-gst-python"

### to set specific gstreamer version
export GST_VERSION=1.14.5

Test

PYTHONPATH=. pytest tests/ -s --verbose

Tools

Setup

• By default Gstreamer tools use libgstreamer-1.0.so.0

export LIB_GSTREAMER_PATH=libgstreamer-1.0.so.0

Export LIB_GSTREAMER_PATH with custom path to libgstreamer.so

Setup Log Level

export GST_PYTHON_LOG_LEVEL=0, 1, 2, 3, 4, 5

Make Gst.Buffer writable

from gstreamer import map_gst_buffer
with map_gst_buffer(pbuffer, Gst.MapFlags.READ | Gst.MapFlags.WRITE) as mapped:
            // do_something with mapped

Make Gst.Memory writable

from gstreamer import map_gst_memory
with map_gst_memory(memory, Gst.MapFlags.READ | Gst.MapFlags.WRITE) as mapped:
            // do_something with mapped

Get Gst.Buffer shape (width,height) from Gst.Caps

from gstreamer import get_buffer_size
ret, (width, height) = get_buffer_size(Gst.Caps)

Convert Gst.Buffer to np.ndarray

from gstreamer import gst_buffer_to_ndarray, gst_buffer_with_pad_to_ndarray

array = gst_buffer_to_ndarray(Gst.Buffer, width, height, channels)
# or
array = gst_buffer_with_pad_to_ndarray(Gst.Buffer, Gst.Pad, channels)

GstPipeline

• With GstPipeline run any gst-launch pipeline in Python

from gstreamer import GstPipeline

command = "videotestsrc num-buffers=100 ! fakesink sync=false"
with GstPipeline(command) as pipeline:
    ...

GstVideoSource based on AppSink

• With GstVideoSource run any gst-launch pipeline and receive buffers in Python

from gstreamer import GstVideoSource

width, height, num_buffers = 1920, 1080, 100
caps_filter = 'capsfilter caps=video/x-raw,format=RGB,width={},height={}'.format(width, height)
command = 'videotestsrc num-buffers={} ! {} ! appsink emit-signals=True sync=false'.format(
num_buffers, caps_filter)
with GstVideoSource(command) as pipeline:
    buffers = []
    while len(buffers) < num_buffers:
        buffer = pipeline.pop()
        if buffer:
            buffers.append(buffer)
    print('Got: {} buffers'.format(len(buffers)))

GstVideoSink based on AppSrc

• With GstVideoSink push buffers in Python to any gst-launch pipeline

from gstreamer import GstVideoSink

width, height = 1920, 1080
command = "appsrc emit-signals=True is-live=True ! videoconvert ! fakesink sync=false"
with GstVideoSink(command, width=width, height=height) as pipeline:
    for _ in range(10):
        pipeline.push(buffer=np.random.randint(low=0, high=255, size=(height, width, 3), dtype=np.uint8))

Metadata

Object Info MedataData

   x
   y
   width
   height
   confidence
   class_name
   track_id

Examples

Run Gstreamer pipeline in Python using Gst.ElementFactory

python examples/pipeline_with_factory.py

Run Gstreamer pipeline in Python using Gst.parse_launch

python examples/pipeline_with_parse_launch.py -p "videotestsrc num-buffers=100 pattern=1 ! autovideosink"

Capture frames (np.ndarray) from any Gstreamer pipeline

PYTHONPATH=. python examples/run_appsink.py -p "videotestsrc num-buffers=100 ! capsfilter caps=video/x-raw,format=RGB,width=640,height=480 ! appsink emit-signals=True"

Push images (np.ndarray) to any Gstreamer pipeline

PYTHONPATH=. python examples/run_appsrc.py -p "appsrc emit-signals=True is-live=True caps=video/x-raw,format=RGB,width=640,height=480 ! queue ! videoconvert ! autovideosink"  -n 1000