Run TFLite on (public) training images in post

Here’s a minimal example that reproduces the issue, extract the attached images to ‘/jpegs’ and run this, hardfaults on first decompress/conversion to RGB565.

import sensor, image, time, os

sensor.reset()
sensor.set_framesize(sensor.QQVGA)
sensor.set_pixformat(sensor.GRAYSCALE)
sensor.set_framebuffers(1)

for jpeg in os.listdir('jpegs'):
    print("Loading:",jpeg)
    image.Image('jpegs/'+jpeg, copy_to_fb=True).to_rgb565()
    time.sleep_ms(3000)

Note, it seems there’s no issue with img.to_grayscale(), which may narrow it down.

With firmware 4.2.3, the same bug still occurs. By the way, I had to manually install it as my IDE isn’t updating by itself.

We did Not fix this issue in v4.2.3, we fixed other issues with TF. And this release wasn’t pushed to IDE resources because we waiting on some updates to the IDE.

Issue for tracking:

This issue has been fixed.

is this included in the current development version? It loaded 4.2.4 and I still have those crashes.

@darrask Yes, you can see the change log in the dev-release notes. I tested all the possible formats and subsampling factors etc… I’m confident it’s fixed, maybe I missed something, please post another minimal example and attach the images.