Skip to content

SeHwanJoo/mmdetection_vinbig

Repository files navigation

top 2% solution

model 1

Detector: Faster RCNN

backbone: ResNext101-32d-4

model 2

Detector: Faster RCNN

backbone: ResNet50

etc: attention

model 3

Detector: Faster RCNN

backbone: RegNet12

model 4

Detector: DetectoRS

backbone: ResNet50

model 5

Detector: DetectoRS

backbone: ResNext101-32d-4

anchor

ratio: [0.5, 1.0, 2.0] -> [0.33, 0.5, 1.0, 2.0, 3.0]

scale: [8] -> [1, 2, 4, 8, 16]

try

Detector: VFNet, Cascade RCNN, Cascade mask RCNN, Hybrid Task Casacde, Generalized Focal Loss

Backbone: EfficientNet, DenseNet

etc: pseudo labeling, segmentation, external data (NIH), Large box detector, small box detector

About

kaggle competition vinbigdata top 2% solution

Resources

License

Code of conduct

Stars

Watchers

Forks

Releases

No releases published

Packages

No packages published

Languages