Jie Hu
Yes, please kindly refer to the doc of Detectron2, which introduces how to implement custom dataset to train your own models.
> SOLOv2 becomes slowest model in your papers, is that really? But as SOLOv2 official paper claims it should faster than most instance segmentation methods, at least compare with MaskRCNN....
Currently, the fastest model is still ISTR-Res50 as I am now lacking of GPUs to train new models with ISTR. (Orz I plan to implement ISTR with mobilenet / vovnet...
@jinfagang yes
> I didn't understand how to position embedding got in this paper. I read this article in detail, but I didn't find the answer. Could you please introduce it? Thank...
Please kindly refer to [Issue #1](https://github.com/hujiecpp/ISTR/issues/1#issuecomment-836039669)
https://github.com/hujiecpp/ISTR/blob/e85d30c1c213355d479054203374f760b8c7f7c4/projects/ISTR/configs/ISTR-AE-R50-3x.yaml
https://github.com/hujiecpp/ISTR/blob/e85d30c1c213355d479054203374f760b8c7f7c4/projects/ISTR/configs/ISTR-AE-R50-3x.yaml
Please kindly refer to README, the model 'link's are listed. Method | backbone | fps | box AP | mask AP | link --- | :---: | :---:|:---: |:---: |:---:...
Pls check if the folder `./detectron2/model_zoo/` exists.