search

marcelsheeny / radiate_sdk

SDK developed to access data from RADIATE dataset
MIT License
94 stars 34 forks source link

How to evaluate the accuracy of a model with a rotation angle on the radiate dataset? #9

Closed dxx111 closed 3 years ago

dxx111 commented 3 years ago

How to evaluate the accuracy of a model with a rotation angle on the radiate dataset?

marcelsheeny commented 3 years ago

Hi,

You should use the DOTA PASCAL Average Precision metric. You can find the repository that evaluates here https://github.com/CAPTAIN-WHU/DOTA_devkit . You need to extract the predicted bounding box values and the ground truth in the format described in the DOTA repository.

dxx111 commented 3 years ago

Okay, thank you very much.

2021-01-23 08:55:25"Marcel Sheeny" notifications@github.com写道：

Hi,

You should use the DOTA PASCAL Average Precision metric. You can find the repository that evaluates here https://github.com/CAPTAIN-WHU/DOTA_devkit . You need to extract the predicted bounding box values and the ground truth in the format described in the DOTA repository.

— You are receiving this because you authored the thread. Reply to this email directly, view it on GitHub, or unsubscribe.

dxx111 commented 3 years ago

Can you share the code of the evaluation model with me?

2021-01-23 08:55:25"Marcel Sheeny" notifications@github.com写道：

Hi,

You should use the DOTA PASCAL Average Precision metric. You can find the repository that evaluates here https://github.com/CAPTAIN-WHU/DOTA_devkit . You need to extract the predicted bounding box values and the ground truth in the format described in the DOTA repository.

— You are receiving this because you authored the thread. Reply to this email directly, view it on GitHub, or unsubscribe.

dxx111 commented 3 years ago

Hello, I tried to use the DOTA toolkit to test the model, the result was lower, I don’t know the reason, can you share your code?

2021-01-23 08:55:25"Marcel Sheeny" notifications@github.com写道：

Hi,

You should use the DOTA PASCAL Average Precision metric. You can find the repository that evaluates here https://github.com/CAPTAIN-WHU/DOTA_devkit . You need to extract the predicted bounding box values and the ground truth in the format described in the DOTA repository.

— You are receiving this because you authored the thread. Reply to this email directly, view it on GitHub, or unsubscribe.

dxx111 commented 3 years ago

I tried to use the DOTA evaluation code evaluation model, the result is still 28.089, I hope you have time to make your evaluation code public, let me learn your code, thank you.

2021-01-23 08:55:25"Marcel Sheeny" notifications@github.com写道：

Hi,

You should use the DOTA PASCAL Average Precision metric. You can find the repository that evaluates here https://github.com/CAPTAIN-WHU/DOTA_devkit . You need to extract the predicted bounding box values and the ground truth in the format described in the DOTA repository.

— You are receiving this because you authored the thread. Reply to this email directly, view it on GitHub, or unsubscribe.

marcelsheeny commented 3 years ago

Can you make sure that you used 0.5 as IoU (Intersection over Union)?

dxx111 commented 3 years ago

Yes, I really use 0.5 as IOU.

2021-02-01 08:03:07"Marcel Sheeny" notifications@github.com写道：

Can you make sure that you used 0.5 as IoU (Intersection over Union)?

— You are receiving this because you authored the thread. Reply to this email directly, view it on GitHub, or unsubscribe.