0
$\begingroup$

I was building a YOLOv5 object detection model, and was looking into researching synthetic methods like GANs to increase the size of my training set in an unsupervised manner.

I know that few-shot GANs can be used to "hallucinate" images and labels for a classification task, but how can they be extended to hallucinate images and labels in YOLO format (basically lists out each bounding box and class)?

Is there some way that I can train a GAN on images / YOLO labels, and get it to hallucinate more images / labels?

$\endgroup$

1 Answer 1

1
$\begingroup$

You can use conditional GANs for synthesizing data with labels/bounding boxes.

Conditional GANs are GANs where, besides the normal random noise (prior), you also insert additional information which has to be taken into account during data generation. You then also pass this same information to the discriminator. The discriminator has to discriminate the real/fake data, taking into account the labels.

Resources: Keras blog about conditional GANs. First ever paper on conditional GANs

$\endgroup$
1
  • $\begingroup$ If the response sufficiently answered your question, please check it as the correct answer using the checkmark so that future visitors know that the question was solved. $\endgroup$ Dec 30, 2022 at 10:34

You must log in to answer this question.

Not the answer you're looking for? Browse other questions tagged .