search

cloud-native-robotz-hackathon / infrastructure

0 stars 1 forks source link

Fedora Detection not precise enough #70

Open nexus-Six opened 3 weeks ago

nexus-Six commented 3 weeks ago

For the model training, increasing the Samples (2000 is too much for the storage), batches and Epochs increases confidence score, but the coordinates always seem to be off. The square is always at the top of the hat.

image

rbo commented 3 weeks ago

Is it really the object detection or the box drawing?

nexus-Six commented 3 weeks ago

The mode is also trained on tennis balls. The coordinates seem to be precise. image

nexus-Six commented 3 weeks ago

This an image from the first trainings before the first workshop, perhaps it was always like that image

nexus-Six commented 1 week ago

Current pipeline default settings provide sufficient detection reliability. Leaving Issue open for some more testing.