FasterRCNN Coco pretrained only predicts human (class 0)

Gluon
1

Hi there,

I am using Yolo, SSD and FRCNN on the same set of images (containing only one object per image, eg car, train etc) to test the differences between models, however, while the other models can correctly determine what is in the image, FRCNN will only detect class 0, or a human.

I’ve used two models both for FRCNN trained on COCO yet they produce the same result. Wondering if anyone could shine some light on this? Wondering if it is characteristic of FRCNN or more likely a bug in my code (however the model always predicts class 0, even as confidence scores change)

Thanks,
David

2

Hi David where are you getting the pretrained models from?

Also for the images that don’t contain any humans, are you saying that the Faster R CNN model makes no detections/predictions at all?

1 Like
3

Hi sad,

Thank you for your offer to help. My apologies, this was due to a bug I had in my code that was well hidden.

This thread can be deleted if a mod sees this.

1 Like