Setting up DOTA for training and evaluation

[Arka161]

I looked into the DOTA website:

https://captain-whu.github.io/DOTA/dataset.html

I see Training , evaluation and testing data. How should I paste these data into the datasets folder for testing BBA? Should I run the DOTA-Devkit too?

I am confused on the set up steps.

Thank you

[yijingru]

https://github.com/yijingru/BBAVectors-Oriented-Object-Detection#about-dota may help

[Arka161]

@yijingru Thank you for your reply.

I currently have datasets/images and datasets/LabelTxt in my directory already. To obtain trainval.txt and test.txt, for running the code, could you suggest how to do it? Thanks

[Arka161]

I see you mention they are the file names, but I am having trouble understanding this block:

P0000__0.5__0___0
P0000__0.5__0___1000
P0000__0.5__0___1500
P0000__0.5__0___2000
P0000__0.5__0___2151
P0000__0.5__0___500
P0000__0.5__1000___0

From my understanding, P0000 is a file name, and I am not sure of the other values

Thank you

[yijingru]

It is image split label. check this one https://github.com/CAPTAIN-WHU/DOTA_devkit#usage

[Arka161]

@yijingru Thank you for all the help.

I managed to set the code base up for testing DOTA! While my own GPU is weak to train from scratch, I am able to perform Inference using your pre-trained model_50.pth file.

There are some small bugs in the readme file, related to running the command for training. For example, the code has num_epochs instead of epochs, and it is not 100% clear what the train_labels and test_labels are, and that we have to do a train-test split using selection.

Can I edit the README.md and make the instructions a little more clear for other researchers via a PR? Thanks.

Either way, I think I am ready to close this issue.

[yijingru]

Of course you can. You can send pull request and I will review and merge. Thanks for your contribution!