Zhihong Fu

> Because the limitation of the BaiduYun disk, can you upload the dataset you mentioned in the `README.MD` to Google Drive? thanks. Limitation of downloading speed.

> I concerned many may have no vpn... Actually, you can provide two addresses, BaiduYun disk and Google Drive.

> 在只改变Batchsize的情况下，当我在一个台 i9-10900k 64GB内存 、3090服务器上使用全数据集训练时，发现内存占用会逐渐上升，并在第13-14个epoch发生内存溢出，这在之前训练时是没有发生过的，而且我除了修改训练配置并未修改任何其他代码，请问是什么原因呢？ 是内存溢出还是显存溢出？训练配置除了batch_size外，还有修改其他内容吗？

如果是内存溢出的话，可以尝试把DataLoader的`pin_memory`参数设置为False。

The hyper-parameter `batch_size` and `start_lr` are important for training. What are your `batch_size` and `start_lr`? `32` and `1e-6` are recommended. If you have difficulty in setting the `batch_size` to `32`...

Can the model converge if you use two GPUs?

Yes. We did this experiment. You can read the section 1.1 of [our supplementary material](https://openaccess.thecvf.com/content/CVPR2021/supplemental/Fu_STMTrack_Template-Free_Visual_CVPR_2021_supplemental.pdf) for more details.

* The role of the position encoding is to add spatial position information to each pixel, it's used by the encoder in both training and inference phase. Refer to the...

> 恭喜前辈发表了这么优秀的论文🎉🎉🎉！ 我是一名新手，经过了两周的研究我终于把项目跑通了，但是现在又遇到了一个问题就是跑出来的结果如何在项目里面进行评估？我上一个跑过的项目可以直接在项目中获得成功率和精度，不知道前辈的这个项目怎么获得 OTB & LaSOT: use the official evaluation tools to get the metrics offline; VOT: directly run the evaluation script in this repo; TrackingNet & GOT-10k: generate trackers'...

"num_segments=4" means that the tracker collects 4 frames from all history frames (except for the first frame and the previous frame). Thus, the number of memory frames in the Space-time...