AI Paper Atlas

Research map / Image generation / Fast diffusion sampling

Diffusion Hyperfeatures: Searching Through Time and Space for Semantic Correspondence

Grace Luo, Lisa Dunlap, Dong Huk Park, Aleksander Holynski, Trevor Darrell

NeurIPS 2023 · Image generation › Fast diffusion sampling · approach: Diffusion models as perception tools · 33 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
Aggregate all layers and timesteps with learned mixing weights instead of a single layer (SD-Layer-4)Stable Diffusion v1-5 features · SPair-71kPCK@0.1 (img): 58.80 -> 72.56Improved
Naively concatenate all layer feature maps instead of learned weighted aggregation (ablation)Stable Diffusion v1-5 features · SPair-71kPCK@0.1 (img): 58.80 -> 52.12Hurt
Use many timesteps (11) instead of a single step (ablation)Diffusion Hyperfeatures · SPair-71kPCK@0.1 (img): 63.74 -> 72.56Improved
Pass the pruned single feature map through the learned bottleneck layer (ablation)Layer 5 Timestep 10 feature map · SPair-71kPCK@0.1 (img): 57.69 -> 64.02Improved
Soft-mix many feature maps instead of pruning to the single best one (ablation)Diffusion Hyperfeatures · SPair-71kPCK@0.1 (img): 64.02 -> 72.56Improved
Swap Stable Diffusion v1-5 for v2-1 as feature source (ablation)Diffusion Hyperfeatures · SPair-71kPCK@0.1 (img): 72.56 -> 70.74Mixed
Increase diffusion timesteps used from 1 to 10 (ablation)Diffusion Hyperfeatures · SPair-71kPCK@0.1 (img): 64.61 -> 72.00Improved
Train aggregation network on DINOv2 features instead of raw layerDINOv2 · SPair-71kPCK@0.1 (img): 60.14 -> 68.37Improved
Concatenate all DINO layers (ablation)DINO · SPair-71kPCK@0.1 (img): 51.68 -> 20.17Hurt
Train aggregation network on DINO featuresDINO · SPair-71kPCK@0.1 (img): 51.68 -> 54.69Improved
Use inversion instead of generation features for real images (ablation)Layer 4 raw features · SPair-71k late timestepsPCK@0.1 (img): up to 5% PCKImproved

Later papers that compared with this method

1 later accepted paper reports Diffusion Hyperfeatures as a baseline in their results tables; 1 of them report a better number on most of the results they share with it.

Matched by method name in results tables read by AI. Authors' own tables tend to favour the newer method, and a comparison can be missed when a method is renamed.

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