Evaluate Rigid Box and Non-Rigid Bags Performance

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

Evaluate Rigid Box and Non-Rigid Bags Performance #18

Open lmtoan opened 5 years ago

lmtoan commented 5 years ago

Replicate what Puru/Havan notebook for bottles.

lmtoan commented 5 years ago

@tejasmhos check this out. The COCO model detected boxes well but failed on bags. Crazy how it recognized the pizza and bowl on the labels. I'll work on batch inference and dissect the predicted class labels (i think right now it only outputs top class prediction)

tejasmhos commented 5 years ago

That's amazing! One option to try out is outputting box whenever it is among the predicted classes, even if it is not the top prediction. I'm working on collecting images for the synthetic generation step for boxes to begin with.