search

IDEA-Research / DINO

[ICLR 2023] Official implementation of the paper "DINO: DETR with Improved DeNoising Anchor Boxes for End-to-End Object Detection"
Apache License 2.0
2.24k stars 250 forks source link

Train the face dataset (wider face) #105

Open xucheng-920 opened 2 years ago

xucheng-920 commented 2 years ago

Hello, thank you for your very beautiful work, I will train the face dataset (wider face), besides modifying num_classes and dn_labelbook_size, what else do I need to modify?

xucheng-920 commented 2 years ago

I have converted the wider face dataset to the format of the coco2017 dataset

SlongLiu commented 2 years ago

I think you can start training a model after modifying num_classes and dn_labelbook_size.

xucheng-920 commented 2 years ago

Thank you for your suggestion, because it is used for face detection tasks, so my target is only faces, I set dn_labelbook_size = 2, num_classes = 2, but the trained model has not been able to detect faces on the validation set.