Shuo Jin

关于BasicVSR++在Vimeo数据集训练的问题

3

@ckkelvinchan 您好，非常感谢您的工作！在我对BasicVSR++进行训练复现时，在REDS数据集上可以得到相应结果。但在Vimeo数据集上，我注意到论文中表述的是" Therefore, we initialize the model using the weights trained on REDS when trained on Vimeo-90K. The number of finetune iterations is 300K"，这里是不是说 在Vimeo数据集上训练，开始时直接load在REDS上训好的模型文件，然后训练300k次iterations就可以了？我之前是这么做的 但效果似乎达不到论文水平 期待您的答复！

About n_segment

2

n_segment是视频序列中帧的个数，但如果不确定视频序列帧数应该怎么办呢，这里不能自适应调整嘛？

Question about object queries

1

Hi, thanks for your impressive work. I'm following fc-clip and find that the number of object queries used in mask2former framework of fc-clip is set as 250, which is usually...

It's been a year since the paper's acceptance. When will the MRES-32M dataset be released? The community can't follow this visual part-grounding work without the released benchmark.