[libcamera-devel] [PATCH v2 10/19] py: examples: Add simple-capture.py

Tomi Valkeinen tomi.valkeinen at ideasonboard.com
Fri May 27 08:08:37 CEST 2022


On 26/05/2022 18:55, Laurent Pinchart wrote:
> Hi Tomi,
> 
> Thank you for the patch.
> 
> On Tue, May 24, 2022 at 02:46:01PM +0300, Tomi Valkeinen wrote:
>> Add an example to showcase the more-or-less minimal capture case.
> 
> Niiiice :-) This is a really good start. I'd like more comments, but
> that can be done on top.
> 
>> Signed-off-by: Tomi Valkeinen <tomi.valkeinen at ideasonboard.com>
>> ---
>>   src/py/examples/simple-capture.py | 153 ++++++++++++++++++++++++++++++
>>   1 file changed, 153 insertions(+)
>>   create mode 100755 src/py/examples/simple-capture.py
>>
>> diff --git a/src/py/examples/simple-capture.py b/src/py/examples/simple-capture.py
>> new file mode 100755
>> index 00000000..45df9e7a
>> --- /dev/null
>> +++ b/src/py/examples/simple-capture.py
>> @@ -0,0 +1,153 @@
>> +#!/usr/bin/env python3
>> +
>> +# SPDX-License-Identifier: BSD-3-Clause
>> +# Copyright (C) 2022, Tomi Valkeinen <tomi.valkeinen at ideasonboard.com>
>> +
>> +# A simple minimal capture example showing:
>> +# - How to setup the camera
>> +# - Capture frames in a blocking manner
>> +# - Memory map the frames
>> +# - How to stop the camera
>> +
>> +import argparse
>> +import binascii
>> +import libcamera as libcam
>> +import libcamera.utils
>> +import sys
>> +
>> +
>> +def main():
>> +    parser = argparse.ArgumentParser()
>> +    parser.add_argument('-c', '--camera', type=str, default='1',
>> +                        help='Camera index number (starting from 1) or part of the name')
>> +    parser.add_argument('-f', '--format', type=str, help='Pixel format')
>> +    parser.add_argument('-s', '--size', type=str, help='Size ("WxH")')
>> +    args = parser.parse_args()
>> +
>> +    cm = libcam.CameraManager.singleton()
>> +
>> +    try:
>> +        if args.camera.isnumeric():
>> +            cam_idx = int(args.camera)
>> +            cam = next((cam for i, cam in enumerate(cm.cameras) if i + 1 == cam_idx))
>> +        else:
>> +            cam = next((cam for cam in cm.cameras if args.camera in cam.id))
>> +    except Exception:
>> +        print(f'Failed to find camera "{args.camera}"')
>> +        return -1
>> +
>> +    # Acquire the camera for our use
>> +
>> +    ret = cam.acquire()
>> +    assert ret == 0
>> +
>> +    # Configure the camera
>> +
>> +    camconfig = cam.generate_configuration([libcam.StreamRole.StillCapture])
> 
> I'm going to bikeshed a bit as this will be used as a starting point by
> many people, sorry.
> 
> As the Python coding style uses snake_case, should this be cam_config ?
> The next variable would be stream_config.
> 
> I'd also use the ViewFinder role to match the C++ simple-cam.

Sounds good to me.

>> +
>> +    streamconfig = camconfig.at(0)
>> +
>> +    if args.format:
>> +        fmt = libcam.PixelFormat(args.format)
>> +        streamconfig.pixel_format = fmt
>> +
>> +    if args.size:
>> +        w, h = [int(v) for v in args.size.split('x')]
>> +        streamconfig.size = libcam.Size(w, h)
>> +
>> +    ret = cam.configure(camconfig)
>> +    assert ret == 0
>> +
>> +    stream = streamconfig.stream
>> +
>> +    # Allocate the buffers for capture
>> +
>> +    allocator = libcam.FrameBufferAllocator(cam)
>> +    ret = allocator.allocate(stream)
>> +    assert ret > 0
>> +
>> +    num_bufs = len(allocator.buffers(stream))
>> +
>> +    print(f'Capturing {num_bufs} frames with {streamconfig}')
> 
> This can be done on top, but I'd like to capture more frames (100 for
> instance), to show how to requeue requests.
> 
>> +
>> +    # Create the requests and assign a buffer for each request
>> +
>> +    reqs = []
>> +    for i in range(num_bufs):
>> +        # Use the buffer index as the cookie
>> +        req = cam.create_request(i)
>> +
>> +        buffer = allocator.buffers(stream)[i]
>> +        ret = req.add_buffer(stream, buffer)
>> +        assert ret == 0
>> +
>> +        reqs.append(req)
>> +
>> +    # Start the camera
>> +
>> +    ret = cam.start()
>> +    assert ret == 0
>> +
>> +    # Queue the requests to the camera
>> +
>> +    for req in reqs:
>> +        ret = cam.queue_request(req)
>> +        assert ret == 0
>> +
>> +    # Wait until the requests are finished
>> +
>> +    reqs = []
>> +
>> +    while True:
>> +        # cm.read_event() blocks until there is an event
>> +        cm.read_event()
>> +
>> +        # Get all the ready requests
>> +        ready_reqs = cm.get_ready_requests()
> 
> Integrating a simple event loop (maybe based on the Python asyncio
> module ?) would be good here.

Let's discuss these in the thread for the following patch.

>> +
>> +        reqs += ready_reqs
>> +
>> +        if len(reqs) >= num_bufs:
>> +            break
>> +
>> +    # Process the captured frames
>> +
>> +    for i, req in enumerate(reqs):
>> +        assert i == req.cookie
>> +
>> +        buffers = req.buffers
>> +
>> +        # A ready Request could contain multiple buffers if multiple streams
>> +        # were being used. Here we know we only have a single stream,
>> +        # and we use next(iter()) to get the first and only buffer.
>> +
>> +        assert len(buffers) == 1
>> +
>> +        stream, fb = next(iter(buffers.items()))
>> +
>> +        # Use MappedFrameBuffer to access the pixel data with CPU. We calculate
>> +        # the crc for each plane.
>> +
>> +        with libcamera.utils.MappedFrameBuffer(fb) as mfb:
>> +            crcs = [binascii.crc32(p) for p in mfb.planes]
> 
> Mapping buffers is expensive, this should be done before starting the
> camera.

Hmm... How do we invalidate the caches if we keep the buffers mapped?

  Tomi


More information about the libcamera-devel mailing list