AI Paper Atlas

Research map / Image generation / Autoregressive and latent generation

D2C: Diffusion-Decoding Models for Few-Shot Conditional Generation

Abhishek Sinha, Jiaming Song, Chenlin Meng, Stefano Ermon

NeurIPS 2021 · Image generation › Autoregressive and latent generation · approach: Hierarchical VQ-VAE generation · 36 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 contrastive SSL (MoCo-v2) loss to the VAE/diffusion latent modelD2C vs NVAE/DDIM · CIFAR-10linear accuracy: 22.5 -> 76.02Improved
add diffusion prior over VAE latents (D2C vs NVAE)D2C · CelebA-64FID: 13.48 -> 5.7Improved
use D2C instead of DDIM for unconditional generationD2C · CIFAR-10FID: 4.16 -> 10.15Hurt
use D2C instead of DDIMD2C · CelebA-HQ-256FID: 25.6 -> 18.74Improved
use D2C latent classifier with rejection sampling instead of DDIMD2C · CelebA-64 few-shot binary MaleFID: 38.38 -> 13.44Improved
use D2C instead of NVAE for few-shot conditional generationD2C · CelebA-64 PU MaleFID: 42.78 -> 16.39Improved
use D2C instead of naive baselineD2C · CelebA-64 binary Non-BlondFID: 3.77 -> 8.94Hurt
fewer diffusion steps (100 -> 10) (ablation)D2C · CIFAR-10FID: 10.15 -> 17.71Hurt
D2C vs StyleGAN2 for image manipulationD2C · CelebA-256 AMTpreference %: 51.5% blond, 60.8% red lipstickImproved
replace StyleGAN2 latent optimization with D2C encoderD2C · manipulation latent timeseconds: 8 seconds -> 0.013 secondsImproved
vary contrastive loss weight lambda (ablation)D2C · unconditionalFIDNo clear effect
train contrastive SSL alone without L_D2 (ablation)contrastive SSL · CIFAR-10linear accuracy: 78.3% vs 76.02Mixed

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