Alternative cloud for trained models

[bobetocalo]

Hello @yangxue0827,

I cannot create a Baidu account. Could you upload the trained models in another cloud such as Dropbox, Onedrive o Google Drive?
If it is not possible ... could you send me all these trained models via email? Thank you so much.

Best regards,
Roberto Valle

[yangxue0827]

I don't have enough space in Google Drive. Which weight do you need, I can send you email.
Please give me the specific configuration file so that I know which weight to download.
example:
cfgs_res50_dota_v4.py
selected from

[bobetocalo]

Thanks @yangxue0827,

I need the three DOTA config files. The most important baselines for me are the following:

RetinaNet-R

cfgs_res50_dota_v7.py
cfgs_res50_dota1.5_v7.py
cfgs_res50_dota2.0_v7.py

FCOS

cfgs_res50_dota_fcos_v1.py
cfgs_res50_dota1.5_fcos_v1.py
cfgs_res50_dota2.0_fcos_v1.py

My email is: bobetocalo@gmail.com

[bobetocalo]

Thank you for the models @yangxue0827

I have some additional questions related to the configuration files.

• Where are the config files of the FCOS algorithm for each DOTA protocol? The links provided are not working:
  https://github.com/yangxue0827/RotationDetection/blob/main/libs/configs/DOTA/fcos/cfgs_res50_dota_fcos_v1.py
  https://github.com/yangxue0827/RotationDetection/blob/main/libs/configs/DOTA1.5/fcos/cfgs_res50_dota1.5_fcos_v1.py
  https://github.com/yangxue0827/RotationDetection/blob/main/libs/configs/DOTA2.0/fcos/cfgs_res50_dota2.0_fcos_v1.py
• How do you import the config files of DOTAv1.5 and DOTA v2.0? It is not easy to import files with dots in the filename ... For example:

from configs.DOTA1.5.retinanet import cfgs_res50_dota1.5_v7

[yangxue0827]

sorry, the relevant code of fcos has been uploaded

import the config files:
Copy $PATH_ROOT/libs/configs/#DATASET/#DETECTOR/cfgs_xxx.py to $PATH_ROOT/libs/configs/cfgs.py

[bobetocalo]

Sorry, the relevant code of fcos has been uploaded

I do not understand. How can I reproduce the results of FCOS using your code? Where are the config files for each DOTA protocol?

FCOS	67.69	61.05	48.10

[bobetocalo]

Sorry, the relevant code of fcos has been uploaded

I do not understand. How can I reproduce the results of FCOS using your code? Where are the config files for each DOTA protocol?

FCOS 67.69 61.05 48.10

Hi @yangxue0827, can I use the FCOS algorithm with the current algorithm? I cannot find the FCOS config files.
On the other hand, I suppose that the aforementioned results in https://github.com/yangxue0827/RotationDetection#latest-performance are obtained using train+val images to train the model and private test annotations to measure the performance... Do you have the results of RetinaNet-R and FCOS using validation images to measure the performance?

Thank you.

[yangxue0827]

config files: https://github.com/yangxue0827/RotationDetection/tree/main/configs/DOTA/fcos

performance in val set:
https://arxiv.org/pdf/2111.06677.pdf