Object detection application giving malloc() error while running

[amararjun]

Object Detection app needs several start and stops before it gets running successfully. It usually takes 8 to 10 times of restarting the app before the malloc() error resolves. I have additional SD card on the camera. Is there any particular reason for this behavior?
I am hoping to have an object detection app that runs on start up and this behavior will most probably make it impossible

[Corallo]

Hello @amararjun

Can you describe more this malloc error that you encounter?

[amararjun]

@Corallo Thank you. I did not explain the situation properly. I am trying MobileNet SSD trained on a custom dataset for object detection. Yes, I have a new label file which has only seven entries ( seven classes). In this situation, can you suggest how to avoid the error? Why is it that the error disappear after few tries ?

[Corallo]

Memory errors are not always consistently reproducible, it could be that it works once in a while when the memory of the device is in a specific state.

Have you updated the Line 13 of the manifest replacing 80 with the number of your classes?
Note, I linked the cpu version of the manifest, if you are using the accelerator, you should it in also the other manifest file

[amararjun]

Thank you for the information.
The 80 in line 13 is supposed to be the image quality, is it not? argparse.c

[Corallo]

Right, my bad, I got confused.

The number of label is actually established by the size of the label file.
So your crash is a bit strange.

Have you made other changes besides the model and the content of the labels file?

[amararjun]

@Corallo Thank you for the cue. A newline was required at the end of the labels.txt file. The malloc() error got solved.