search

yijingru / BBAVectors-Oriented-Object-Detection

[WACV2021] Oriented Object Detection in Aerial Images with Box Boundary-Aware Vectors
MIT License
466 stars 87 forks source link

testing issue #85

Closed minsu-kim320 closed 3 years ago

minsu-kim320 commented 3 years ago

Thank you for sharing the code. I have one question. The test datasets are cropped into 600×600 patches with a stride of 100 and two scales 0.5 and 1. (using /BBAVectors-Oriented-Object-Detection/datasets/DOTA_devkit/SplitOnlyImage.py)

image

The total number of testset(testsplit) images is 15394. When tested on the dota task1 evaluation server(DOTA-v1.0 Server) using the trained model posted on the Googledrive (with confidence threshold is 0.1):

model_50.pth

mAP: 0.3793444297656531 ap of each class: plane:0.5374595665861279, baseball-diamond:0.27002700270027, bridge:0.4343089364781523,ground-track-field:0.1466275659824047,small-vehicle:0.250634814569665-ball-field:0.17947177744823833, roundabout:0.1662971175166297, harbor:0.669322425400183, swimming-pool:0.644582662489113, helicopter:0.45122344393356406

model_43.pth

mAP: 0.3798813073173642 ap of each class: plane:0.5363196690278388, baseball-diamond:0.26886150153476884, bridge:0.4328021217790057, ground-track-field:0.14171122994652408, small-vehicle:0.2531614365ccer-ball-field:0.2044285046000312, roundabout:0.1590909090909091, harbor:0.6642627607270606, swimming-pool:0.6341358688145124, helicopter:0.4917699838529368

I wonder how I can solve it.

Thank you.

yijingru commented 3 years ago

As described in paper input images have two scales 0.5 and 1. The trainval set and testing set contain 69,337 and 35,777 images after the cropping, respectively. It looks like you don't have 0.5 scale images in the testing set. By the way, please submit merge_result to the evaluation server.

minsu-kim320 commented 3 years ago

Thank you for your reply I solved the problem.