AI Paper Atlas

Research map / 3D vision / 3D generation

Pri3D: Can 3D Priors Help 2D Representation Learning?

Ji Hou, Saining Xie, Benjamin Graham, Angela Dai, Matthias Nießner

ICCV 2021 · 3D vision › 3D generation · approach: Semantic scene completion · 68 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 Pri3D pre-training (view+geo) on ScanNet after ImageNet initResNet50 U-Net · ScanNet semantic segmentationmIoU: 55.7 -> 61.7 (+6.0)Improved
add view-invariant contrastive loss (ablation)Pri3D ResNet50 · ScanNet semantic segmentationmIoU: 55.7 -> 61.3Improved
add geometric prior contrastive loss (together with view loss) (ablation)Pri3D ResNet18 · ScanNet semantic segmentationmIoU: 54.4 -> 55.7Improved
add geometric prior to view loss (ablation)Pri3D ResNet50 · NYUv2 semantic segmentationmIoU: 54.2 -> 54.7No clear effect
add ScanNet data to MoCoV2 without 3D priors (combine)MoCoV2 ImageNet · ScanNet semantic segmentationmIoU: 55.7 -> 54.9Hurt
add ScanNet data to MoCoV2 by mixing batchesMoCoV2 ImageNet · ScanNet semantic segmentationmIoU: 55.7 -> 54.5Hurt
replace trivial color-augmentation correspondences with multi-view 3D correspondences (ablation)Pri3D ResNet50 · ScanNet semantic segmentationmIoU: 56.4 -> 61.3Improved
replace single-frame depth prediction pre-training with Pri3D view+geo (ablation)ResNet50 · ScanNet semantic segmentationmIoU: 58.4 -> 61.7Improved
add depth proxy loss (ablation)Pri3D (View) · ScanNet semantic segmentationmIoU: 60.2 -> 61.3Improved
add Pri3D to DeepLabV3DeepLabV3 ResNet50 · ScanNet semantic segmentationmIoU: 57.0 -> 61.3Improved
add Pri3D to PSPNetPSPNet ResNet50 · ScanNet semantic segmentationmIoU: 59.7 -> 62.8Improved
add Pri3D to ImageNet init on NYUv2 detectionResNet50 · NYUv2 object detectionAP@0.5: 29.9 -> 34.0Improved

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