YuHengsss
commented
7 months ago Thank you for your interest in our work.
YOLOX is trained from scratch, which is why it requires 300 epochs on the COCO dataset. In contrast, our video detector is initialized with a base detector that has been fine-tuned on the VID dataset, and most of the base detector's weights are frozen during this process. This approach significantly reduces the number of training epochs (iterations) required. The 7 epochs we used is an empirical setting based on our experiments. If you wish to train on your own dataset, you can use this as an initial setting and adjust it as necessary based on your specific requirements.
We hope this clarifies your query. Please feel free to reach out if you have any further questions.
Thank you for your excellent work. I noticed that the maximum number of iterations during YOLOX training is 300, while your YOLOV maximum number of iterations is 7. Why do you set it this way?, If I want to train my dataset, how should I set up the training epochs？