AI Paper Atlas

Research map / 3D vision / 3D generation

3x2: 3D Object Part Segmentation by 2D Semantic Correspondences

Anh Thai, Weiyao Wang, Hao Tang, Stefan Stojanov, James M Rehg, Matt Feiszli

ECCV 2024 · 3D vision › 3D generation · approach: Semantic scene completion

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
generate non-overlapping masks from SAM proposals (ablation)3-By-2 2D segmentation · PartNetE few-shot (Scissors, Mouse, Suitcase, Bottle, Chair)mIoU: 0.457 -> 0.675 (Scissors), 0.004 -> 0.810 (Bottle)Improved
aggregate multi-view labels per mask instead of per point (ablation)3-By-2 3D aggregation · PartNetE few-shot (Scissors, Suitcase, Printer, Clock)mIoU: 0.363 -> 0.458 (Clock), 0.009 -> 0.085 (Printer)Improved
replace VLPart 2D segmenter with SAM plus DIFT label transfer (ablation)3-By-2 with mask-consistency aggregation · PartNetE zero-shot (PACO)mIoU: 0.222 -> 0.430Improved
add more categories to the label database (ablation)3-By-2 · PartNetE KettlemIoU: 0.815 -> 0.757 -> 0.752Hurt
add Chair objects to Table database for wheel transfer (ablation)3-By-2 · PartNetE tables with wheelsmIoU: 0.411 -> 0.625 average; wheel 0.000 -> 0.600Improved
increase labeled instances in the database from 1 to 8 (ablation)3-By-2 · PartNetE KettlemIoU: 0.809 -> 0.815No clear effect
increase number of rendered views beyond 10 (ablation)3-By-2 · PartNetEmIoUNo clear effect
use training-free label transfer instead of language-guided GLIP detectionPartSLIP · PartNet level-3mIoU: 0.125 -> 0.281Improved
3-By-2 versus MvDeCor on fine-grained PartNetMvDeCor · PartNet level-3mIoU: 0.277 -> 0.281No clear effect
sparse random pixel sampling for label transfer across runs (ablation)3-By-2 · PartNetE (Scissors, Kettle, Suitcase, Bottle)mIoU std: std 0.002 to 0.008No clear effect
coarse-to-fine correspondence search (ablation)3-By-2 · 800x800 image pairwall time speedup: approximately 2000 timesImproved

Earlier methods this paper compares with

Named baselines from earlier accepted papers in this paper's results tables: better than 1 of 1 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