How to use Background images in training?

Accelerated Computing Intelligent Video Analytics TAO Toolkit
, ,
1

How to use Background images in training? just add Background images into training images or add Background images into training images and add corresponding empty Kitti labels into training labels?

The basic reason to add background images is to reduce unwanted FP detection in the random background area.

I wanted to train using yolo_v4 with tao-toolkit.

3

It is not working for adding empty Kitti labels. For your case, please generate more training images which have more objects in the random background area.

1 Like
4

I Didn’t get your point exactly. Sorry.
I’m a training model for a single class only. Can you explain to me, how I should add more objects in random background areas if I’m having only one class for training? & In the training dataset, I have at least a minimum of 3 objects per image to a maximum of 16 objects/image.

If possible please share the answer with more details.
thank you.

5

There is no update from you for a period, assuming this is not an issue anymore.
Hence we are closing this topic. If need further support, please open a new one.
Thanks

The empty images which have no objects are not supported to used for training.
So, to improve the mAP when train with current images, you can

  • Double check the anchor shapes if they are correct
  • Use deeper backbone
  • Finetune lr, bs, etc.
  • If possible, use Imagenet pretrained model

Then, you can also add more training images. Please make sure the new training images have the objects.

7

This topic was automatically closed 14 days after the last reply. New replies are no longer allowed.