AI Paper Atlas

Research map / Detection and segmentation / CNN features

Towards Rotation Invariance in Object Detection

Agastya Kalra, Guy Stoppi, Bradley Brown, Rishav Agarwal, Achuta Kadambi

ICCV 2021 · Detection and segmentation › CNN features · approach: Box regression and loss design · 17 citations (OpenAlex)

Read the paperCheck your idea against 137,217 papers

AI-extracted. The measured effects below were read by AI from the paper's full text (no summary yet). Numbers and claims can be wrong or missing context, so check the paper before you cite or build on it.

Measured effects reported in the paper

What was changedWhereMeasuredEffect
replace Largest Box labels with Ellipse + RU Loss for rotation augmentationobject detector (Detectron2) · Transparent Object Bin PickingAP75: largest box almost 50% worse, ours 4.5% betterImproved
use Ellipse label instead of Largest Box (and every other prior tried) (ablation)rotation augmentation for COCO detection · COCO val2017 rotated 0-40 degreesAPImproved
add RU Loss with delta=10 degrees on top of Ellipse labels (ablation)Ellipse label rotation · COCO val2017 rotated 0-40 degreesAPImproved
use larger delta in RU Loss on simpler datasets than COCO (ablation)RU Loss · simpler detection datasetsAPImproved
higher Expected IoU of the rotation label method (ablation)label rotation methods · COCO val2017 at 10 degreesAPImproved
use Ellipse versus other label methods at 0 degrees (ablation)rotation label methods · COCO val2017 at 0 degreesAPNo clear effect
use COCO shape priors instead of random shapes for the optimization (ablation)Ellipse method · COCOAPNo clear effect

Closest papers by meaning

Writing a paper in this area? scime draft is a free LaTeX editor with AI included, real-time co-authors and a check of your references and venue rules before you submit. It also imports OpenAI Prism projects.

Open scime draft

Get the next scime Dispatch and a monthly field update

Free. We send a confirmation link first; every email has a one-click unsubscribe. Privacy