Feng Li
No. `diff` is a matrix that defines the largest noise that can be added to each corrdinates. `diff[:, 2:]` is the `w, h` noise, which is the same as the...
Yes, these parameters are set without considering the batch size. If you reduce batch size, you are expected to increase the steps accordingly.
You can use `instances.gt_boxes = instances.gt_masks.get_bounding_boxes()` instead.
Hi, thanks for this question. We have updated the twiter data several days ago, please download a new one. For the data in MagicBrush, we will check the data and...
Hi, for `image_aspect_ratio`, you can set to `anyres`. For single-image data, we do anyres training. But do not use anyres for multi-image data to avoid excessive computation cost.
DINO is not designed to support segmentation, therefore we cannot guarantee its performance. Please wait for the code release of [Mask DINO](https://github.com/IDEA-Research/MaskDINO) that extends DINO to three segmentation tasks.
Hi, thanks for reaching out. The deviation is a bit large. Did you use the `mask-guided anchor box initialization`? Please clone our latest code and see if the setting is...
Great! We would appreciate it if you could contribute to this repo with the hugging face hub. Thanks!
Hi, `anyres` is only used for single-image inference in our training and evaluation, because using anyres for multi-image will significantly increase the training cost. Computational cost is the main concern,...
Not yet. We will release it in the future.