AI Paper Atlas

Research map / Video / Video generation

Diff-Foley: Synchronized Video-to-Audio Synthesis with Latent Diffusion Models

Simian Luo, Chuanhao Yan, Chenxu Hu, Hang Zhao

NeurIPS 2023 · Video › Video generation · approach: Video-to-audio synthesis · 16 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 CAVP visual features instead of ResNet50 (ablation)Diff-Foley · VGGSoundAlign Acc (%): 54.88 -> 92.35Improved
use CAVP visual features instead of CLIP (ablation)Diff-Foley · VGGSoundAlign Acc (%): 79.23 -> 92.35Improved
use CAVP features instead of CLIP (ablation)Diff-Foley · VGGSoundInception Score: 57.43 -> 52.07Hurt
add Temporal Split & Merge Augmentation (ablation)Diff-Foley · VGGSoundAlign Acc (%): 79.79 -> 92.35Improved
increase model size from 335M to 859M (no pretrain) (ablation)Diff-Foley · VGGSoundInception Score: 16.96 -> 24.91Improved
initialize with pretrained SD-V1.4 weights (ablation)Diff-Foley-L · VGGSoundInception Score: 24.91 -> 52.07Improved
add classifier-free guidance (ablation)Diff-Foley · VGGSoundAlign Acc (%): 69.41 -> 92.35Improved
add classifier guidance alone (no CFG) (ablation)Diff-Foley · VGGSoundAlign Acc (%): 69.41 -> 66.05Hurt
add classifier guidance on top of CFG (double guidance) (ablation)Diff-Foley · VGGSoundAlign Acc (%): 92.35 -> 94.78Improved
pretrain CAVP on VGGSound plus AudioSet-V2A instead of VGGSound only (ablation)Diff-Foley · VGGSoundAlign Acc (%): 85.88 -> 92.35Improved
use Diff-Foley instead of Im2Wavvideo-to-audio · VGGSoundAlign Acc (%): 67.40 -> 94.05Improved
use Diff-Foley instead of SpecVQGAN/Im2Wav on distribution metricsvideo-to-audio · VGGSoundFID/KL: FID 8.93 (SpecVQGAN) vs 9.87; KL 5.20 (Im2Wav) vs 6.43Mixed

Later papers that compared with this method

11 later accepted papers report Diff-Foley as a baseline in their results tables; 11 of them report a better number on most of the results they share with it.

3 more not listed.

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