Crash with DFK 33UX273 Camera at 200 FPS and 1000 Exposure, Works Fine with Basler QCAM-UC1440-220CE

[bulverismo]

Hi!

Describe the bug When I use a camera The Imaging Source DFK 33UX273 with 200 fps and exposure 1000 my application crashes, but when I use a Basler QCAM-UC1440-220CE camera it doesn't crash

To Reproduce I set the camera to 1000 us of exposure and set the fps to 200

Expected behavior I hope to get the images for later processing with any camera

Camera description: CAMERA THAT OCCURS THE BUG

• Manufacturer The Imaging Source
• Model DFK 33UX273
• Interface USB3

CAMERA THAT DOESN'T OCCUR THE BUG

• Manufacturer Basler
• Model DFK QCAM-UC1440-220CE
• Interface USB3

Platform description:

• Aravis version 0.8.31
• OS: Debian GNU/Linux 12 (bookworm)
• Hardware Intel Atom(R) x6413E Processor @ 1.50GHz - 8G Memory

Additional context

capture loop

` def __non_triggered_capture(self): with self.lock_var_to_update: for i in range(1, 10): buffer = self._stream.timeout_pop_buffer(10000 * 1000) if buffer: if not (buffer.get_status().value_name == 'ARV_BUFFER_STATUS_SUCCESS'): continue logging.info(self._stream.get_statistics()) img = self._converter(buffer) timestamp = buffer.get_timestamp() self._stream.push_buffer(buffer) self.frame_id = (self.frame_id + 1) % (sys.maxsize - 1) if img is None: raise CameraError("Camera capture failed") return img, self.frame_id, timestamp / 1000000 logging.warning(f"Failed capturing from camera {self.camera_id}, model {self.dev}. Retrying {10 - i} times") raise CameraError("Camera capture failed") `

Log the imaging source Logs basler

[EmmanuelP]

                  img = self._converter(buffer)
                  if img is None:
                      raise CameraError("Camera capture failed")

Capture failed is raised when the call to converter() return a None. I don't know what converter() is supposed to do.

[bulverismo]

It is a function to convert img in raw format to numpy array

def _converter(self, buf):
        if not buf:
            return None
        pixel_format = buf.get_image_pixel_format()
        bits_per_pixel = pixel_format >> 16 & 0xff
        # Determine the appropriate ctypes data type based on bits per pixel
        if bits_per_pixel == 8:
            INTP = ctypes.POINTER(ctypes.c_uint8)
            bytes_per_pixel = 1
        elif bits_per_pixel == 16:
            INTP = ctypes.POINTER(ctypes.c_uint16)
            bytes_per_pixel = 2
        elif bits_per_pixel == 24:
            INTP = ctypes.POINTER(ctypes.c_uint8)
            bytes_per_pixel = 3
        elif bits_per_pixel == 32:
            INTP = ctypes.POINTER(ctypes.c_uint8)
            bytes_per_pixel = 4
        else:
            raise ValueError(f"Unsupported bits per pixel: {bits_per_pixel}")

        addr = buf.get_data()
        ptr = ctypes.cast(addr, INTP)

        height = buf.get_image_height()
        width = buf.get_image_width()

        # Determine the number of channels based on bytes per pixel
        if bytes_per_pixel == 1:
            im = np.ctypeslib.as_array(ptr, (height, width))
        else:
            im = np.ctypeslib.as_array(ptr, (height, width, bytes_per_pixel))

        # Copy the image to ensure it is not a view into the original buffer
        im = im.copy()

        return im

[EmmanuelP]

Thanks.

At some point in the log, the payload size returned in the image trailer packet changes from 4665600 to 187396. Do you change some features between 2 call to __non_triggered_capture ?

[bulverismo]

Yes, after initializing the camera the application starts capturing(using __non_triggered_capture) and then during the capture the application configure some features, such as exposure, gain and gamma, for example.

[EmmanuelP]

exposure, gain and gamma

Is this an exhaustive list ? Any feature that may modify the payload size ?

[bulverismo]

Full list: frame rate region exposure time exposure auto gain gamma balance ration balance white auto trigger activation trigger delay trigger

I think the only part that can change the payload size is the part where I configure the image size(which I called region in the list above).

set window size method

        # makes the x's and y's a multiple of 4
        x0 -= x0 % 4
        x1 -= x1 % 4
        y0 -= y0 % 4
        y1 -= y1 % 4

        self.cam.stop_acquisition()
        self.cam.set_region(
            x0,  # offset x
            y0,  # offset y
            max(x1 - x0, 16),  # 16*8 pixel minimal image
            max(y1 - y0, 8),
        )

        self.create_buffer()
        self.cam.start_acquisition()

create buffer method

def create_buffer(self):

    if hasattr(self, '_stream'):
        del self._stream

    self._stream = self.cam.create_stream(None, None)
    payload = self.cam.get_payload()
    for _ in range(0, 10):
        self._stream.push_buffer(Aravis.Buffer.new_allocate(payload))

[EmmanuelP]

You should also stop the acquisition thread when you want to set the region, otherwise there will still be buffer with the wrong size in the queues. Please have a look at tests/arvroitest.c, switch_roi() function.