Skip to content

Latest commit

 

History

History

mmdet-nwdrka

README.md

Installation and Get Started

Required environments:

Install TODbox:

Note that our TODbox is based on the MMDetection 2.24.1. Assume that your environment has satisfied the above requirements, please follow the following steps for installation.

git clone https://github.com/Chasel-Tsui/mmdet-aitod.git
cd mmdet-nwdrka
pip install -r requirements/build.txt
python setup.py develop

Get Started

Train a network with with single GPU, for example, Faster R-CNN w/ NWD-RKA:

python tools/train.py configs_nwdrka/nwdrka/aitod_faster_r50_nwdrka_1x.py

Performance

Table 1. Training Set: AI-TOD-v2 trainval set, Validation Set: AI-TOD-v2 test set, 12 epochs

Method Backbone mAP AP50 AP75 APvt APt APs APm
FR R-50 12.8 29.9 9.4 0.0 9.2 24.6 37.0
DR R-50 16.1 35.5 12.5 0.1 12.6 28.3 40.0
FR w/ NWD-RKA R-50 21.4 53.2 12.5 7.7 20.7 26.8 35.2
DR w/ NWD-RKA R-50 24.7 57.4 17.1 9.7 24.2 29.3 39.3

FR denotes Faster R-CNN, DR denotes DetectoRS

For your convenience, we also provide the performance of the model trained on AI-TOD-v2 train set and validated on the AI-TOD-v2 val set. Table 2. Training Set: AI-TOD-v2 train set, Validation Set: AI-TOD-v2 val set, 12 epochs

Method Backbone mAP AP50 AP75 APvt APt APs APm
FR R-50 12.9 29.5 9.2 0.0 9.5 27.3 37.2
FR w/ NWD-RKA R-50 21.9 51.8 13.9 5.8 21.8 27.3 37.8

Table 3. Training Set: DOTA-v2 train set, Validation Set: DOTA-v2 val set, 12 epochs, HBB Task

Method Backbone mAP AP50 AP75 APvt APt APs APm
FR R-50 35.6 59.5 37.2 0.0 7.1 28.9 42.1
DR R-50 40.8 62.6 44.4 0.0 7.0 29.9 47.8
FR w/ NWD-RKA R-50 36.4 61.5 37.6 1.5 10.4 29.4 43.2
DR w/ NWD-RKA R-50 41.9 66.3 44.4 1.9 10.6 30.3 48.5

Please refer to the paper for detailed performance on the AI-TOD, AI-TOD-v2, DOTA-v2 and VisDrone2019.