AI Paper Atlas

Research map / Image generation / GANs

DreamTeacher: Pretraining Image Backbones with Deep Generative Models

Daiqing Li, Huan Ling, Amlan Kar, David Acuna, Seung Wook Kim, Karsten Kreis, Antonio Torralba, Sanja Fidler

ICCV 2023 · Image generation › GANs · approach: GAN-based image manipulation · 19 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 DreamTeacher feature distillation (ADM) pretraining instead of supervised ImageNet pretrainingResNet-50 backbone · ADE20k segmentationmIoU: 40.9 -> 42.5Improved
use ADM as teacher instead of Stable Diffusion 1.4 or eDiffI (ablation)DreamTeacher RN50 · BDD100K segmentationmIoU: 61.6 vs 60.2 (SD1.4) vs 58.5 (eDiffI)Improved
DreamTeacher pretraining vs BYOL on in-domain nuScenes-30KBEVFormer RN50 · nuScenes 3D detection valNDS: 22.2 -> 28.4Improved
DreamTeacher pretraining vs BYOL on ImageNet-1kBEVFusion RN50 camera only · nuScenes map segmentationMean IoU: 55.4 -> 53.9Hurt
DreamTeacher pretraining vs supervised on camera+LiDAR fusionBEVFusion RN50 · nuScenes map segmentation C+LMean IoU: 59.2 -> 59.4No clear effect
increase diffusion noise step used for features (ablation)DDPM-seg FFHQ · Sim2real MS-SynFaces-19mIoU: 61.5 -> 44.2Hurt
use a few diffusion steps instead of step 0 (ablation)DDPM-seg LSUN-car · Car-20mIoU: 64.4 -> 75.4Improved
use DT mixed distillation with ConvNeXt-B instead of DDPM-seglabel-efficient segmentation · Bedroom-28mIoU: 49.4 -> 54.8Improved

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