wanghao9610
commented
2 months ago Hello, @kaijinz0228 , thank you for your interest in our work. We don't have enough computational resources to conduct more experiments, while it is almost certain that OV-DINO will obtain better results with more training data or a larger model size. We have established O365 pre-training code, you could pre-train it on the O365 v2 dataset.
Dear authors,
Have you conducted some experiments to evaluate the scalability of OV-DINO, like using objects365-v2 to train the model？The Grounding DINO-L model was trained using the v2 version.
Thank you