AI Paper Atlas

Research map / Image generation / Autoregressive and latent generation

DIVER: Diving Deeper into Distilled Data via Expressive Semantic Recovery

Qianxin Xia, Zhiyong Shu, Wenbo Jiang, Jiawei Du, Jielei Wang, Guoming Lu

ICML 2026 · Image generation › Autoregressive and latent generation · approach: Latent diffusion with learned tokenizers

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 DIVER stage IIMTT · ImageFruit IPC=10 unseen architecturesaccuracy (%): 18.9 -> 29.8Improved
add DIVER stage IISRe2L · ImageNet-1K IPC=50 ResNet-18accuracy (%): 46.8 -> 54.0Improved
add DIVER stage II as squeeze model grows strongerSRe2L / G-VBSM · ImageNet-1K ResNet-101accuracy (%): 61.0 -> 63.7Mixed
add DIVER to MGD3 prototypesMGD3 · ImageNet-1K IPC=10accuracy (%): 45.8 -> 46.4Improved
add Minimax fine-tuning to MGD3+DIVERMGD3 + DIVER · ImageNet-1K IPC=10accuracy (%): 46.4 -> 46.9Improved
add semantic inheritance (ablation)DIVER · ImageFruit IPC=10, MTTaccuracy (%): 23.4 -> 26.2Improved
add semantic guidance (ablation)DIVER · ImageFruit IPC=10, MTTaccuracy (%): 23.4 -> 27.8Improved
add semantic guidance to semantic inheritance (ablation)DIVER · ImageFruit IPC=10, MTTaccuracy (%): 26.2 -> 28.3Improved
add semantic fusion (guidance only in semantic phase) (ablation)DIVER · ImageFruit IPC=10, MTTaccuracy (%): 28.3 -> 29.8Improved
use raw DiT without recovery modules (ablation)DiT · ImageFruit IPC=10, MTTaccuracy (%): 18.9 -> 23.4Improved
replace distilled images with random images in DIVER (ablation)DIVER · ImageFruit IPC=10, MTTaccuracy (%): 29.8 -> 21.7Hurt
swap SD-V1.5 for DiT as diffusion model (ablation)DIVER · ImageNette IPC=10, MTTaccuracy (%): 26.5 -> 34.3Improved

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