AI Paper Atlas

Research map / 3D vision / 3D generation

Improving 2D Feature Representations by 3D-Aware Fine-Tuning

Yuanwen Yue, Anurag Das, Francis Engelmann, Siyu Tang, Jan Eric Lenssen

ECCV 2024 · 3D vision › 3D generation · approach: Semantic scene completion · 35 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
concatenate original DINOv2 features with 3D-aware fine-tuned features instead of adding them (ablation)FiT3D · NYUv2 segmentationmIoU: 66.0 -> 67.5Improved
concatenate instead of linear fusion (ablation)FiT3D · NYUv2 segmentationmIoU: 66.39 -> 67.5Improved
increase per-Gaussian feature dimension from 32 to 64 (ablation)feature Gaussians · NYUv2 segmentationmIoU: 67.15 -> 67.5Improved
fine-tune for 2 or 3 epochs instead of 1 (ablation)FiT3D · NYUv2 segmentationmIoU: 67.5 -> 67.25No clear effect
fine-tune DINOv2 directly instead of frozen DINOv2 plus adapter (ablation)FiT3D · Pascal VOC segmentationmIoU: 82.02 -> 82.35No clear effect
add fine-tuned 3D-aware features to DINOv2 (vs just duplicating original features) (ablation)linear probing · ScanNet++ segmentationmIoU: 30.31 -> 32.76Improved
apply 3D-aware fine-tuning to DINOv2 ViT-SDINOv2 · ScanNet++ segmentationmIoU: 30.19 -> 32.76Improved
apply 3D-aware fine-tuning (trained only on ScanNet++) to out-of-domain depthDINOv2 · KITTI depthRMSE: 3.03 -> 2.91Improved
apply 3D-aware fine-tuning to other backbones (CLIP, MAE, DeiT-III, DINOv2-reg)2D vision models · ScanNet++ segmentationmIoU: MAE 17.19 -> 20.27Improved
apply 3D-aware fine-tuning to DINOv2DINOv2 · ImageNet-1K classificationaccuracy: 80.02 -> 80.00No clear effect
use ViT-B instead of ViT-S for 3D-aware fine-tuningDINOv2 · ScanNet++ segmentationmIoU: 32.72 -> 36.19Improved
use DPT head instead of linear probe with 3D-aware featuresDINOv2 · ScanNet++ depthRMSE: 0.3027 -> 0.2830Improved

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