An

> @ANdong-star Hi, this process is quite similar to that in the DETR decoder. In DETR, 100 object queries interact with the image features output by the encoder. In STARK,...

> Hi, thanks for your work！ Could you pleae provide the Hz value of GOT-10K when evaluating the Stark? Oh, I got it. I find many trackers in the Leaderboard...

> @ANdong-star Hi, the reason is that we evaluate STARK with 32 threads by default. This would slow down the tested speed of each sequence, but save time in total....

> @ANdong-star Hi, the metric of these two ways may not have difference. But we found that training with "bs16 on 8GPUs" is slightly faster than "bs32 on 4 GPUs",...

> @ANdong-star Hi, you can refer to the visualization code of [DETR NoteBook](https://colab.research.google.com/github/facebookresearch/detr/blob/colab/notebooks/detr_attention.ipynb) :) ok, really thanks!