-
Where did you get the camera module(s)?
kickstarter preorder -
Model number of the product(s)?
hawkeye 64mp -
What hardware/platform were you working on?
rpi4 4gb -
Instructions you have followed. (link/manual/etc.)
Followed the PDF manual provided by Arducam -
Problems you were having?
I’m capturing large datasets by successively setting the manual focus and callinglibcamera-still -n -o tmp.jpg --width 9152 --height 6944
. Every so often, an image is showing a duplicate of a previous image in a set, but not the most recent previous image. Duplicates confirmed both visually and by md5.
As an example, I might capture 5 images, with the first 4 being unique, and the 5th image being a duplicate of the 1st image.
This is also periodically happening in cycles. In a segment of 60 captures, images 1 through 20 are unique, images 21-40 are copy of images 1-20, and images 41-60 are also a duplicate of images 1-20.
Notably, this issue does not occur with a 16mp camera (also from Arducam), and it also doesn’t occur when running this 64mp camera at 16mp resolution.
-
The dmesg log from your hardware?
I did not think to capture this at the time, and these take a while to generate, as the dataset capture takes multiple hours. I will start another capture and make sure to grab it this time. -
Troubleshooting attempts you’ve made?
Increasing gpu-mem (didn’t help), adding a significant delay between image captures (didn’t help), adding dispmanx_offline=1 to /boot/config.txt (made the problem less frequent, but not entirely solved) -
What help do you need?
Someone with a better understanding of how low-level memory management works on the rpi to help narrow down which buffers are being incorrectly recycled.
Relevant (changed) parts of /boot/config.txt:
[all]
camera_auto_detect=0
gpu_mem=256
dtoverlay=vc4-kms-v3d,cma-512
dtoverlay=imx519,media-controller=1
dtoverlay=arducam_64mp,media-controller=0
dispmanx_offline=1