search

TannerGilbert / MaskRCNN-Object-Detection-and-Segmentation

Train your own custom MaskRCNN Object Detection and Instance Segmentation model.
https://gilberttanner.com/blog/getting-started-with-mask-rcnn-in-keras
56 stars 24 forks source link

Detection score greater than 0.9 for wrong detections #2

Closed vkramdev closed 3 years ago

vkramdev commented 3 years ago

Hi Gilbert,

I am using matterport implementation of maskrcnn. I have trained my model on my custom dataset , however during inference if the instance is not detected correctly in an image ,say a ball is detected as knife even in this case the score shown with ball is 0.978 .

How do i reduce this score so that it conveys better message say the instance is detected as knife however the score is something like 0.5 or 0.4.

What should i be looking into to resolve this . Any help or suggestion is appreciated.

Thanks in advance.

TannerGilbert commented 3 years ago

Maybe getting more training images or using data augmentation could help

github-actions[bot] commented 3 years ago

This issue is stale because it has been open 30 days with no activity. Remove stale label or comment or this will be closed in 5 days