icvteam / tdrg Goto Github PK

The work is quite interesting!
Can the checkpoint of this work be released?

Dear author,

Thanks very much for interesting work and code!!
I have tried to reproduce the reported results using the original code and settings described in the paper, but I can not reproduce the reported results on both COCO and VOC.

For COCO, reported: 84.6 (448;448), reproduced: 80.5 (448;448)
For VOC, reported: 95.0 (448;448), reproduced: 94.0 (448;448)

Can you help me solve this issue?
Thanks very mch for your time!!

Very solid work. Figure 2 in the paper is beautifully done, both in layout and color scheme. Can you release the source file for Fig2? I want some of the material of Fig2 to draw my own figure.

Hi,

Awesome work! I'm curious to know if you can share the activation map or visualization script.

thanks for your detailed codes. I wonder about the ablation study settings of Fig.3 . Since SGD optimizer largely depends on the scale of the loss, is it necessary to multiple a single loss (for example, trans loss) by four to match the initial loss scale when I want to validate the effect of trans loss?

File "main.py", line 6, in
from trainer_baseline import Trainer
ModuleNotFoundError: No module named 'trainer_baseline'

How do I get this ?

请问数据集失效了吗，点击链接不能跳转。感谢作者。