-
我看到这篇论文在推送中说被收录到CVPR2021,但是我去谷歌上搜索没看到被收录
-
why don't you use lite-hrnet as your backbone? limited performence or lite-hrnet isn't suitable in this paper
-
Hi,
is it possible to comment on how you obtained the dataset splits, in the specific case of CUB. You mention you follow the configuration of [53] (_Good, bad an ugly_), but there are 3 different…
-
Hi,
In `mode/model_proto.py, line 76-78`, the file for loading pretrained backbone weights is **resnet101_c.pth.tar**
My question is, what is the difference between **resnet101_c.pth.tar** an…
-
Dear author!
There are some related papers of CVPR2021.
I copy this from https://github.com/52CV/CVPR-2021-Papers#19
![image](https://user-images.githubusercontent.com/28667932/117594008-cd5f6780…
ghost updated
3 years ago
-
Please add our Oral **Domain Generalisation** paper, "Generalization on Unseen Domains via Inference-time Label-Preserving Target Projections"
Link: https://openaccess.thecvf.com/content/CVPR2021/p…
-
-
Hello, is there any difference between your code and BMN series? According to my understanding, this task should be the same label format as activitynet1.3. In addition, your dataset is 32x of Activit…
-
Can you share your link to CVPR2021 paper?
-