NiklasKroeger-AlliedVision
commented
3 years ago Are you certain that you want to use cv2.imwritemulti? I have been playing around with it and from my understanding it is intended to support multi-page Tiff files and nothing else. I personally have not been able to find an image viewer that easily allows me to view these multi-page Tiff files.
To tackle the problem you are describing (imwrite is too slow to be used in the frame callback) I have a different suggestion: create a thread that is responsible for writing out your recorded frames. This way the frame callback does not have to perform the slow disk-io. Instead it only takes the resulting opencv image from the received frame and puts it into a queue from which the thread writing to disk takes them. of course this queue will keep growing if you continuously record new images and these images are recorded faster than they can be written to disk, but if it is only intended to record a certain number of frames I believe this may solve your problem. Here is a code example doing just that:
import functools
import queue
import threading
import time
import cv2
import vimba
class ImageRecorder:
def __init__(self, cam: vimba.Camera, frame_queue: queue.Queue):
self._cam = cam
self._queue = frame_queue
def __call__(self, cam: vimba.Camera, frame: vimba.Frame):
# Place the image data as an opencv image and the frame ID into the queue
self._queue.put((frame.as_opencv_image(), frame.get_id()))
# Hand used frame back to Vimba so it can store the next image in this memory
cam.queue_frame(frame)
def _setup_software_triggering(self):
# Always set the selector first so that folling features are applied correctly!
self._cam.TriggerSelector.set('FrameStart')
# optional in this example but good practice as it might be needed for hadware triggering
self._cam.TriggerActivation.set('RisingEdge')
# Make camera listen to Software trigger
self._cam.TriggerSource.set('Software')
self._cam.TriggerMode.set('On')
def record_images(self, num_pics: int):
# This method assumes software trigger is desired. Free run image acquisition would work
# similarly to get higher fps from the camera
with vimba.Vimba.get_instance():
with self._cam:
try:
self._setup_software_triggering()
self._cam.start_streaming(handler=self)
for i in range(num_pics):
print(i)
self._cam.TriggerSoftware.run()
# Delay between images can be adjusted or removed entirely
time.sleep(0.1)
finally:
self._cam.stop_streaming()
def write_image(frame_queue: queue.Queue):
while True:
# Get an element from the queue.
frame, id = frame_queue.get()
cv2.imwrite(f'image_{id}.png', frame)
# let the queue know we are finished with this element so the main thread can figure out
# when the queue is finished completely
frame_queue.task_done()
def main():
num_pics = 60
with vimba.Vimba.get_instance() as vmb:
cams = vmb.get_all_cameras()
frame_queue = queue.Queue()
recorder = ImageRecorder(cam=cams[0], frame_queue=frame_queue)
# Start a thread that runs write_image(frame_queue). Marking it as daemon allows the python
# program to exit even though that thread is still running. The thread will then be stopped
# when the program exits
threading.Thread(target=functools.partial(write_image, frame_queue), daemon=True).start()
recorder.record_images(num_pics=num_pics)
frame_queue.join()
if __name__ == "__main__":
main()I hope this helps. If you have further questions or if this does not solve the problem please let me know.
frischwood
newtop95
I have a question related to the last framework posted above: Is there a nice and easy way (preferably built in vimba) to stack the 10 frames (and many more optionally) from the software triggering loop and save them as a batch afterwards using
cv2.imwritemulti()for example? And if not, how to make sure each frame can get written before the next Software triggering happens?I ask this because I'm having problems at saving the images with
cv2.imwrite()within theframe_callback(). The Triggering goes faster than the writing. In the end several images are missing on disk.Just to give you an idea it looks as follows:
Also I'd prefer to remove the 1 sec. delay after each triggering. Many thanks for the material already shared. It helped a lot!!
Originally posted by @frischwood in https://github.com/alliedvision/VimbaPython/issues/60#issuecomment-883559785