Diffusion meets Object Detection
DiffusionDet: Diffusion Model for Object Detection
• It formulates object detection as a denoising diffusion process
from noisy boxes to object boxes
• At training, object boxes diffuse from ground-truth boxes to random distribution
• The model learns to reverse this noising process
• During inference, the model gradually refines a set of randomly generated boxes to refine the results
• DiffusionDet has better COCO metric scores compared to many classical object detection models
abs: https://arxiv.org/abs/2211.09788
pdf: https://arxiv.org/pdf/2211.09788.pdf
source code: https://github.com/ShoufaChen/DiffusionDet
@farid Very cool new object detection paradigm!
Though it takes a bit of a hit in terms of FPS, this will probably be improved in a few follow-up papers by the community.
Comes with great flexibility of using the model at evaluation time.
@lb I agree with you, Lucas.
I look forward to seeing more papers on object detection/segmentation using diffusion.
@farid This is really great, IMO applying this kind of iterative methods to new types of object detection, for example those described by graphs or curves will be super interesting.