AI Paper Atlas

Research map / Point clouds / Point cloud classification

KeypointDETR: An End-to-End 3D Keypoint Detector

Hairong Jin, Yuefan Shen, Jianwen Lou, Kun Zhou, Youyi Zheng

ECCV 2024 · Point clouds › Point cloud classification · approach: Local feature aggregation networks · 10 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
use bipartite-matching multi-heatmap end-to-end detection instead of shared heatmap with NMS/clusteringKeypointDETR · KeypointNet saliency (avg)Hungarian mIoU (%): 82.70 -> 93.81 vs Wei et al.Improved
add query-wise transformer decoder instead of MLP decoders (ablation)KeypointDETR · KeypointNet saliency (avg)Hungarian mIoU (%): 91.67 -> 93.81Improved
add position encoding in decoder (ablation)KeypointDETR · KeypointNet saliency (avg)Hungarian mIoU (%): 92.47 -> 93.81Improved
use heatmap-based prediction instead of coordinate-based prediction (ablation)KeypointDETR · KeypointNet saliency (avg)Hungarian mIoU (%): 49.83 -> 93.81Improved
add cascade offset prediction to coordinate-based variant (ablation)KeypointDETR coordinate-based · KeypointNet saliency (avg)Hungarian mIoU (%): 49.83 -> 58.81Improved
use dynamic graph transformer encoder instead of DGCNN encoder (ablation)KeypointDETR · KeypointNet saliency (avg)Hungarian mIoU (%): 89.81 -> 93.81Improved
use dynamic graph transformer encoder instead of PointNet++ or Point Transformer encoder (ablation)KeypointDETR · KeypointNet saliency (avg)Hungarian mIoU (%): 73.28 -> 93.81 (PN2); 80.45 -> 93.81 (PT)Improved
replace encoder with pre-trained Point-BERT, Point-MAE or I2P-MAE (ablation)KeypointDETR · KeypointNet saliency (avg)Hungarian mIoU (%): 92.88 / 90.88 / 83.57 vs 93.81No clear effect
use KeypointDETR instead of Wei et al. for correspondenceKeypointDETR · KeypointNet correspondence (avg)Hungarian mIoU (%): 80.12 -> 87.09Improved

Earlier methods this paper compares with

Named baselines from earlier accepted papers in this paper's results tables: better than 2 of 2 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