luogen1996

> enviroment: > pytorch 1.5.1 > cuda 10.1 > test on small input tensor (2,8,5,5) > > when using the test method in lib/sa/functions to test the speed, I found...

> Hi, thank you for your work! I want to know how you get your anchors for the datasets? Could you share the code or anything else? I did not...

Thank you for pointing out this typo error. The descriptions in the paper might be 'umd' instead of 'unc'. In fact, there are 104,560 expressions in the original RefCOCOg (google),...

> Thank you for sharing this code! > > I am testing your code for multitask video with BART on 24GB GPUs. To run your code on 24GB GPUs, I...

> I just found out that DDP works well with full fine-tuning but works worse with parameter-efficient transfer learning methods. I will further investigate this issue soon. If DDP does...

yes, i found the same problems. Meanwhile, i think the shape of `self.relative_x` should be `(groups,self.rel_size, self.kernel_size[1])` instead of `(self.rel_size, self.kernel_size[1])` . In fact, when i fixed this problems, the...

I think this implements are incorrect which ignores the effect of "multi-head". In fact, the implements are single-head, thereby making it more effecient, but less effectively. For example, when you...

Dimensions are set to 8, the same as that in Table 1.

This function is borrowed from [here](https://github.com/microsoft/Swin-Transformer/blob/f92123a0035930d89cf53fcb8257199481c4428d/main.py#L277), which is a common practice in image classification. I guess that it is used to optimize the utilization of the GPU cache through multiple...

Source codes for [video classification](https://github.com/luogen1996/RepAdapter/tree/main/video_task) and [image segmentation](https://github.com/luogen1996/RepAdapter/tree/main/segmentation_task) are updates in our repository.