Cardboard Box detection

I have done some training using DetectNet v2 and have gotten ok results using vgg 16 as the base network.

I have a dataset of cardboard boxes, which base network do you recommend to use; it seems that the ones most people are getting good results are using Resnet 18/34?

Do you have a recommendation on a detection architecture other then detectnetv2?

In TLT 2.0 , six detection networks are available.
detectnet_v2
ssd
faster-rcnn
dssd
retinanet
yolo

And each detection network supports several backbones, see tlt 2.0 user guide for more details.

For your case, it is needed to trigger more experiments to train your own dataset.
And also need to consider your target for the mAP or FPS.

For higher fps, I think you can try yolo or ssd too.

1 Like

Hi ishan, is it possible to share your dataset of cardboard boxes? I am trying to do this exact same project. I would be willing to pay for your help if need be.

Hey ishan,

how did you generate your cardboard boxes dataset?