issues
search
ViTAE-Transformer
/
RSP
The official repo for [TGRS'22] "An Empirical Study of Remote Sensing Pretraining"
MIT License
133
stars
6
forks
source link
issues
Newest
Newest
Most commented
Recently updated
Oldest
Least commented
Least recently updated
ViTAE_Window_NoShift 这个文件夹里面实现的是ViTAEv2吗
#21
WenLinLliu
opened
1 month ago
1
speed up training by replacing sklearn with torchmetrics
#20
Winter-Jon
closed
8 months ago
0
semantic segmentation pretraining weight
#19
SCM556
closed
8 months ago
0
关于MillionAID数据集
#18
yyq0828
opened
8 months ago
1
new datasets
#17
zhu011
opened
9 months ago
1
what is the version of mmcv and mmseg
#16
dotlla
opened
9 months ago
1
关于semantic segmentation 测试结果可能会有差异
#15
youngbaldy
opened
10 months ago
7
训练时间+GPU是否利用问题
#14
WeiCL7777
opened
11 months ago
4
是否可以提供MAE的代码或相关参考
#13
AnRanbel
closed
11 months ago
1
upernet_vit_base_win.py model config is in the Remote-Sensing-RVSA repo
#12
utkutpcgl
closed
11 months ago
1
使用levir数据集报错问题
#11
wangqingl
opened
12 months ago
3
你好,我训练的时候报错如下图,可以看一下吗
#10
wangqingl
opened
1 year ago
12
您好,请问你们使用的ISAID数据集是原始数据集吗?
#9
mobaigy
opened
1 year ago
3
你好,RSP-ResNet-50-E300给出的模型是训练好的权重吗?
#8
lucky85712
opened
1 year ago
10
When model's pretrained is none, get AttributeError: 'ViTAE_Window_NoShift_basic' object has no attribute 'use_abs_pos_embed'
#7
AnRanbel
closed
1 year ago
4
huge_img_inference patch coordinates bug fix
#6
nezihkasim
opened
1 year ago
1
Runtime error
#5
mobaigy
opened
1 year ago
1
Classification train/test set split.
#4
zcablii
opened
1 year ago
1
Warning: grad and param do not obey the gradient layout contract. This is not an error, but may impair performance.
#3
AnRanbel
closed
1 year ago
1
is there still have pretrained weights saved for the SeCo dataset
#2
AnRanbel
closed
1 year ago
1
Change Detection requirements
#1
luzhiyuan787
closed
1 year ago
1