search

Ranchosky / OAN

Apache License 2.0
53 stars 7 forks source link

About OAN's attempt in horizontal box object detection #7

Open XiaoyuZxy00 opened 1 year ago

XiaoyuZxy00 commented 1 year ago

Dear Esteemed Author,

I am currently endeavoring to train my own horizontal box dataset using the OAN. However, upon reviewing your source code, I have noticed that the implementation of OAN is primarily based on the rotated box DOTA dataset.

I would like to ask if you have achieved object detection on the horizontal box (such as coco dataset)? Alternatively, I would appreciate any guidance you can offer on how I could modify the relevant configurations to adapt the OAN for use with horizontal box object detection.

I am looking forward to your response and greatly appreciate your help in this matter.

zhangpeng2001 commented 1 year ago

@XiaoyuZxy00 You can reference oan_obb_two_stage.py(OAN/mmdet/models/detectors/obb/oan_obb_two_stage.py) to modify your model for horizontal bounding box. The component, OAN, is defined in utils(OAN/mmdet/models/utils/oan.py). This structure of this project is similar with MMDetection of Open-MMLab, you can refer to it (https://github.com/open-mmlab/mmdetection).