FPN_Tensorflow icon indicating copy to clipboard operation
FPN_Tensorflow copied to clipboard

Published 20 hours ago • verified publisher icon yangxue0827

Results are not good for dota dataset

Open heiyuxiaokai opened this issue 6 years ago • 5 comments

The dataset is dota of whu These images are cropped into (1200 x 1200) by step 900, about 8000 images. 1.The total mAP is about 0.38 when the global step is 80000, and some classes can not be detected, such as basketball_court, ground_track_field, soccer_ball_field 2.Total loss is declining from 8 to 0.4 when the global step is 260000(it seems to be the limit of this code?). But the mAP is 0.34 and some classes that can be detected in step 80000 can not be detected in step 260000. I do some changes in the train.py for multi_gpus_train, and I have met the same problem when I use the origin code of one gpu.

heiyuxiaokai avatar Dec 03 '18 03:12 heiyuxiaokai

Recommend improved code: https://github.com/DetectionTeamUCAS/FPN_Tensorflow. @heiyuxiaokai

yangxue0827 avatar Dec 03 '18 03:12 yangxue0827

@heiyuxiaokai When you train the DOTA dataset, have you changed the parameters inside the cfgs.py? What are the specific changes? Thank you.

WhiteSheep250 avatar Jul 13 '19 14:07 WhiteSheep250

Reference there: https://github.com/DetectionTeamUCAS/FPN_Tensorflow/issues/59. @WhiteSheep250

heiyuxiaokai avatar Jul 14 '19 04:07 heiyuxiaokai

Ok, I will try that method. Thank you. @heiyuxiaokai

WhiteSheep250 avatar Jul 14 '19 06:07 WhiteSheep250

Does the new method fit the DOTA dataset?

WhiteSheep250 avatar Jul 14 '19 06:07 WhiteSheep250