open-mmlab
Swin V2
Right now, we do not support swin v2, but you can use swin. Examples are here: https://github.com/open-mmlab/mmdetection/tree/master/configs/swin. swin v2 is in the planning and will be supported soon
I have seen swin v2 implemnted in mmcls.
- Have any swinv2 shedule in mmdet?
- Can i directly copy the code from mmcls to mmdet?
Right now, we do not support swin v2, but you can use swin. Examples are here: https://github.com/open-mmlab/mmdetection/tree/master/configs/swin. swin v2 is in the planning and will be supported soon
Is swin v2 ready now?
Right now, we do not support swin v2, but you can use swin. Examples are here: https://github.com/open-mmlab/mmdetection/tree/master/configs/swin. swin v2 is in the planning and will be supported soon When will the swin v2 be got in mmdet? When I try to let mmdet support swin V2, I got a bad performance. I don't know the reason. Are there some modifications that should be done?
I have seen swin v2 implemnted in mmcls.
Have any swinv2 shedule in mmdet?
Can i directly copy the code from mmcls to mmdet?
did you try? did it work?
I have seen swin v2 implemnted in mmcls.
- Have any swinv2 shedule in mmdet?
- Can i directly copy the code from mmcls to mmdet?
did you try? did it work?
I haven't pay attention on this issue for a long time. Back then, I had a try on this (directly copy the code). In my case, downsampling resolution does't fit my project. However, if I changed the model structure, the pre-trained weight would not work and I was unable to train it. Had a try to train my model from scratch, I got a pretty bad result. Then I didn't move on.
https://mmpretrain.readthedocs.io/en/latest/papers/swin_transformer_v2.html
(v2 here for anyone interested, I might see if I can get any decent results porting it over)