The roll, yaw and pitch of strawberries.

My arxiv frontpage found another fun article about a new benchmark. This one is about orientation estimation of strawberries. It turns out that this is super useful for fruit picking robots.

The picture from the article really helps explain it.

A few things were interesting here. One of them was the architecture for keypoint detection. I was well aware of the UNET architecture, but these researches have gone with a variant that they call "hourglass".

Resnet-y vibes.

You might wonder what keypoints are needed. There's a few, but the top and tip of the strawberry are the main ones.

Resnet-y vibes.

It's a fun read, the only shame is that I couldn't find any links to the strawberry datasets that they used for benchmarking.