search

havanagrawal / clomask

Capstone Project for Clobotics: Using Mask R-CNN for Rigid/Non-Rigid Retail Consumable Product Detection
MIT License
7 stars 4 forks source link

Pre train results #24

Open pshivraj opened 5 years ago

pshivraj commented 5 years ago

This PR adds Inference notebook for preliminary pre-trained models result on validation data.

lmtoan commented 5 years ago

Thanks Puru.

For the image with falling bottles, I think we achieved the objective of fixing the alignment recognition. But for other images, there seems to be a lot of false positives in the background or on container labels. How does mAP metric penalize false positives?

Since this method is working (overfitting) for bottles, I have a lot of hopes that we can improve performance on cereal box and chip bags dramatically. Right now, COCO pre-trained does not do so well.