Qikai Li

> How long did you take to get the best result? and how many epochs did it take to converge to the best result?

> hi, you can modify the attention layer, the current code does not include that Thanks for your quick reply! Could you please offer me some help to modify the...

> yes, you can use nn.MultiheadAttention, it is similar to the usage of cross attention in detr, with query as the joint query, key and value as the multi-view image...

I have tried the above setting and didn't get results near `MPJPE 114.5mm`. So I was wondering i**f there are any fields in the config I should also modify** to...

Here is the result i got after 74 epochs: ``` 2022-04-14 21:17:41,738 Test: [0/323] Time: 239.402s (239.402s) Speed: 0.0 samples/s Data: 238.601s (238.601s) Memory 455115776.0 2022-04-14 21:18:14,152 Test: [100/323] Time:...

The above is the best result I got. And I tried to turn the lr to `1e-5` but it didn't get better. > the full attention result is not got...

How did you get your result of Table 5 in your paper? Do you have any suggestions that I should change to make my result approach yours? 

1&4. Both can. You can refer to this: https://mmpose.readthedocs.io/en/latest/inference_speed_summary.html 2. For high-resolution image, you can detect the instances in the image and use the bboxes to crop the image and...

I got the same problem as well when I reimplement the algorithm. The training speed is really slow and the performance is really bad. Does anyone have a solution?

@GXYao7 Hi. Thanks for your feedback. The result seems abnormal. Could you please provide more information? For example, the config file you used, the command you run, and the possible...