pppppM

We found that the file size of the three different checkpoint sizes is the same, but the parameter size in each checkpoint is different. This seems to be some problem...

Sorry, we don't know the point rend model very well and can't give very accurate recommendations. At present, there is no very general KD algorithm that can be applied to...

Sorry, we have the same problem as you. The DDP mechanism of OpenMMLab is not friendly enough to the unconventional training process. The developers of MMCV are refactoring this problem.

The final performance may be related to the model and data you use, but I have also encountered similar experimental phenomena as you. Question 2 may require more information so...

I think the idea you mentioned should be feasible, `DetNAS` does it in detection. However, we have not done similar experiments due to our limited human resources. If you are...

We haven't found an efficient distillation method on yolox yet, if you want to develop a distillation algorithm on yolox, this document may be helpful to you. https://github.com/open-mmlab/mmrazor/blob/master/docs/en/tutorials/Tutorial_6_customize_kd_algorithms.md

Sorry for replying to you so late. It seems to be caused by the different results of the ROI bbox head, not a bug. CWD loss requires two inputs of...

Sorry, we haven't found a good algorithm for yolox yet.

@Yaolang003 Using DetNAS in distillation model means distilling the searched model by DetNAS ?

This represents where you want to calculate the distillation loss in the model. If you modify the teacher or student, you also need to modify the module name accordingly.