AI Paper Atlas

Research map / Detection and segmentation / CNN features

Cascade R-CNN: High Quality Object Detection and Instance Segmentation

Zhaowei Cai, Nuno M. Vasconcelos

TPAMI 2019 · Detection and segmentation › CNN features · approach: Region-based and query-based detectors · 2,077 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
add cascade of detectors with increasing IoU thresholds (Cascade R-CNN)FPN+ ResNet-50 · COCO val AP90AP90: 7.1 -> 15.8Improved
add iterative BBox regression at inferenceFPN+ baseline · COCO valAP: 34.9 -> 35.4No clear effect
add integral loss with three classifiersFPN+ baseline · COCO valAP90: 7.1 -> 6.9No clear effect
increase IoU threshold across stages (ablation)Cascade R-CNN · COCO valAP: 36.8 -> 38.5Improved
update regression statistics per stage (ablation)Cascade R-CNN · COCO valAP: 38.5 -> 38.9Improved
use avg stage loss weighting instead of decay (ablation)Cascade R-CNN · COCO valAP90: 15.8 -> 16.2Mixed
add a second stage (ablation)FPN+ baseline · COCO valAP: 34.9 -> 38.2Improved
add a third stage (ablation)2-stage Cascade R-CNN · COCO valAP: 38.2 -> 38.9Improved
add a 4th stage (u=0.75) (ablation)3-stage Cascade R-CNN · COCO valAP: 38.9 -> 38.6No clear effect
add cascade to detectors across architecturesFaster R-CNN, R-FCN, FPN+ · COCO test-devAP: 38.8 -> 42.8Improved
add mask head to the last stage rather than the first (ablation)Cascade Mask R-CNN · COCO val instance segmentationmask AP: 35.0 -> 35.4Improved
add a mask branch to every stage (ablation)Cascade Mask R-CNN · COCO val instance segmentationmask AP: 35.4 -> 35.5No clear effect

Earlier methods this paper compares with

Named baselines from earlier accepted papers in this paper's results tables: better than 3 of 3 on most shared results.

Matched by method name in results tables read by AI. Authors' own tables tend to favour the newer method, and a comparison can be missed when a method is renamed.

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