Can't set Camera.still.setSize higher than what Camera.video allows

DerChie

I'm trying to migrate to camera-node instead of color/mono-camera with my OAK D Lite
When trying to do full-res saving via still-out, using following code:

FW is on 1.0.0

import depthai as dai
import cv2

pipeline = dai.Pipeline()

# Define sources and outputs
camRgb: dai.node.Camera = pipeline.create(dai.node.Camera)

#Properties
camRgb.setBoardSocket(dai.CameraBoardSocket.CAM_A)
camRgb.setSize((4208, 3120))

# Linking

stillOut = pipeline.create(dai.node.XLinkOut)
stillOut.setStreamName("still")
camRgb.still.link(stillOut.input)

with dai.Device(pipeline) as device:
    still = device.getOutputQueue(name="still", maxSize=1, blocking=False)

    while True:
        if still.has():
            cv2.imshow("still", still.get().getCvFrame())
        if cv2.waitKey(1) == ord('q'):
            break

I'm retrieving following error:

RuntimeError: Camera(0) - 'video' width or height (4208, 3120) bigger than maximum at current sensor resolution (3840, 2160)

Shouldn't I actually be able to set the resolution higher than what's max possible with the video-stream? I'm not even trying to use it.

OskarSonc

Hey @DerChie - You’re not doing anything wrong – with camera-node setSize() still configures the video stream, which is capped at 3840×2160, so 4208×3120 trips that check even if you only use still.
For DepthAiv3 we have this script:
https://docs.luxonis.com/software-v3/depthai/examples/camera/camera_still_max_res/

DerChie

Ok, does it mean, there is no possibility to capture a max sensor resolution still image with the camera node in v2?

OskarSonc

It should work on v2 too, but we don't have ready to run example for that, here are docs:
https://docs.luxonis.com/software/depthai-components/nodes/camera/

DerChie

Apparently it doesn't...
Using camera.setStillSize() as in following code

#!/usr/bin/env python3
import depthai as dai
import cv2 as cv
import time

pipeline = dai.Pipeline()

# Define sources and outputs
cam_rgb: dai.node.Camera = pipeline.create(dai.node.Camera)

#Properties
cam_rgb.setBoardSocket(dai.CameraBoardSocket.CAM_A)
cam_rgb.setStillSize((4208, 3120))

# I/O
control_in = pipeline.create(dai.node.XLinkIn)
still_out = pipeline.create(dai.node.XLinkOut)


# Linking
## RGB
cam_rgb.still.link(still_out.input)
control_in.out.link(cam_rgb.inputControl)

# Streamnames
control_in.setStreamName('control')
still_out.setStreamName('still')

with dai.Device(pipeline) as device:
    print("Camera initialized and ready to use.")
    control_queue = device.getInputQueue('control', maxSize=1, blocking=False)
    still_queue = device.getOutputQueue('still', maxSize=20, blocking=False)
    ctrl = dai.CameraControl()
    ctrl.setCaptureStill(True)
    control_queue.send(ctrl)
    time.sleep(2)

    color_frame = still_queue.get().getCvFrame()
    cv.imshow('still', color_frame)

Retrieves

RuntimeError: Camera(0) - 'still' width or height (4208, 3120) bigger than maximum at current sensor resolution (1920, 1080)

Looks very wild to me, because now i don't even have 4k resolution available...
For some reason this seems super weird to me because the ColorCamera-node accepts the following code, with my png file as output:

#!/usr/bin/env python3
import depthai as dai
import cv2 as cv
import time

pipeline = dai.Pipeline()

# Define sources and outputs

cam_rgb = pipeline.create(dai.node.ColorCamera)

#Properties
cam_rgb.setBoardSocket(dai.CameraBoardSocket.CAM_A)
#cam_rgb.setStillSize((4208, 3120))
cam_rgb.setResolution(dai.ColorCameraProperties.SensorResolution.THE_13_MP)

# I/O
control_in = pipeline.create(dai.node.XLinkIn)
still_out = pipeline.create(dai.node.XLinkOut)

# Linking
## RGB
cam_rgb.still.link(still_out.input)
control_in.out.link(cam_rgb.inputControl)

# Streamnames
control_in.setStreamName('control')
still_out.setStreamName('still')

with dai.Device(pipeline) as device:
    print("Camera initialized and ready to use.")
    control_queue = device.getInputQueue('control', maxSize=1, blocking=False)
    still_queue = device.getOutputQueue('still', maxSize=20, blocking=False)
    ctrl = dai.CameraControl()
    ctrl.setCaptureStill(True)
    control_queue.send(ctrl)
    time.sleep(2)

    color_frame = still_queue.get().getCvFrame()
    cv.imwrite("test_image.png", color_frame)

OskarSonc

Okay, I did a bit of digging and got it working. DaiV2 docs aren't that complete and there are some not well explained stuff, since we moved to v3...

I got it working by adding this:
cam.setSize(4056, 3040)
cam.setStillSize(4056, 3040)
cam.setVideoSize(3840, 2160)

I tested this on OAK-D-PRO which has IMX378 that's why I set resolution of 4056x3040.
I think you should be able to set it to 4208x3120.

This is the whole script:

#!/usr/bin/env python3
import cv2 as cv
import depthai as dai
import time

pipeline = dai.Pipeline()

cam = pipeline.create(dai.node.Camera)
cam.setBoardSocket(dai.CameraBoardSocket.CAM_A)
cam.setSize(4208, 3120)       # sensor output
cam.setStillSize(4208, 3120)  # still stream size
cam.setVideoSize(3840, 2160)  # video stream size

# IO
control_in = pipeline.create(dai.node.XLinkIn)
control_in.setStreamName("control")
control_in.out.link(cam.inputControl)

still_out = pipeline.create(dai.node.XLinkOut)
still_out.setStreamName("still")
cam.still.link(still_out.input)

with dai.Device(pipeline) as device:
    print("Camera initialized and ready to use.")
    control_q = device.getInputQueue("control", maxSize=1, blocking=False)
    still_q = device.getOutputQueue("still", maxSize=20, blocking=False)

    ctrl = dai.CameraControl()
    ctrl.setCaptureStill(True)
    control_q.send(ctrl)
    time.sleep(2)

    frame = still_q.get().getCvFrame()
    cv.imwrite("test_image.png", frame)
    cv.imshow("still", frame)
    cv.waitKey(0)

Hope this helps,
Oskar