AI Paper Atlas

Research map / 3D vision / 3D generation

NOVUM: Neural Object Volumes for Robust Object Classification

Artur Jesslen, Guofeng Zhang, Angtian Wang, Wufei Ma, Alan Yuille, Adam Kortylewski

ECCV 2024 · 3D vision › 3D generation · approach: 2D-to-3D pretrained representations · 13 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 3D Gaussian cuboid volume with a single feature vector per class (ablation)NOVUM · P3D+ L0 and occluded-P3D+ L3classification accuracy (%): 99.5 -> 93.2 (L0); 59.2 -> 44.0 (L3)Hurt
add background features during training (ablation)NOVUM · occluded-P3D+ L3classification accuracy (%): 53.0 -> 59.2Improved
add background features during training (ablation)NOVUM · P3D+ L0 (IID)classification accuracy (%): 99.3 -> 99.5No clear effect
use cuboid instead of sphere Gaussian layout (ablation)NOVUM · occluded-P3D+ L3classification accuracy (%): 59.0 -> 59.2No clear effect
use neural object volumes instead of feed-forward classifiersNOVUM · OOD-CVclassification accuracy (%): 64.2 -> 85.2 (vs Swin-T)Improved
use neural object volumes instead of feed-forward classifiersNOVUM · occluded-P3D+ meanclassification accuracy (%): 75.8 -> 81.6 (vs Convnext)Improved
use neural object volumes instead of feed-forward classifiers on unoccluded dataNOVUM · P3D+ L0classification accuracy (%): 99.4 -> 99.5No clear effect
use NOVUM volume rendering instead of NeMo mesh render-and-compare for poseNOVUM · P3D+ 3D pose pi/6accuracy (%): 86.1 -> 88.2Improved
use NOVUM instead of NeMo for fine pose thresholdNOVUM · P3D+ 3D pose pi/18accuracy (%): 61.0 -> 59.1Mixed
use class-agnostic backbone and neural volumes instead of per-class NeMo modelsNOVUM · pose inference efficiencyparameters / optimization steps: 996M -> 83M params; 300 -> 30 stepsImproved
increase parameter count among baselinesclassification baselines · OOD robustnessaccuracy: Swin-T 28M, ViT-b-16 86M, NOVUM 83MNo clear effect

Later papers that compared with this method

2 later accepted papers report NOVUM as a baseline in their results tables; 1 of them report a better number on most of the results they share with it.

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