fcjian / InstaGen

InstaGen: Enhancing Object Detection by Training on Synthetic Dataset, CVPR2024
https://fcjian.github.io/InstaGen
MIT License
68 stars 2 forks source link

Which novel categories are included in the synthetic images used for training in Open Vocabulary Object Detection? #10

Open Bilibilee opened 1 month ago

Bilibilee commented 1 month ago

On the open-vocabulary COCO benchmark, do the synthetic images include the 17 novel categories in COCO-OVD? If that is the case, would it be considered as leaking the novel categories?