AI Paper Atlas

Research map / Vision-language and multimodal / Human and embodied modeling

Towards Effective Multi-Modal Interchanges in Zero-Resource Sounding Object Localization

Yang Zhao, Chen Zhang, Haifeng Huang, Haoyuan Li, Zhou Zhao

NeurIPS 2022 · Vision-language and multimodal › Human and embodied modeling · approach: Region-level vision-language alignment · 0 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
replace real query with a random vector in the localization stream (ablation)TURN · VGGSScIoU@0.5: 12.2 -> 34.6Improved
replace real query with a random vector in the localization stream (ablation)TURN · MUSIC-SoloIoU@0.5: 16.9 -> 33.7Improved
replace real query with a random vector in the localization stream (ablation)TURN · MUSIC-SoloAUC: 34.1 -> 45.2Improved
use RefCOCOg instead of RefCOCO as source domain (AudioCaps intermediate) (ablation)TURN · VGGSScIoU@0.5: 23.13 -> 33.58Improved
use RefCOCOg instead of RefCOCO+ as source domain (Clotho intermediate) (ablation)TURN · VGGSScIoU@0.5: 22.51 -> 34.57Improved
use Clotho instead of AudioCaps as intermediate domain (RefCOCOg source) (ablation)TURN · VGGSScIoU@0.5: 33.58 -> 34.57Improved
use AudioCaps instead of Clotho as intermediate domain (RefCOCOg source) (ablation)TURN · MUSIC-SoloIoU@0.5: 28.21 -> 33.68Improved
remove contrastive loss (largest drop among component ablations) (ablation)TURN · VGGSScIoU@0.5 (Clotho): 34.57 -> 26.44Improved
remove codebook loss, reconstruction loss or adaptive sampling (one at a time) (ablation)TURN · VGGSScIoU@0.5 (Clotho): 34.57 -> 28.56Improved
vary total codebook size (too large or too small) (ablation)TURN codebook · VGGSScIoU and AUCMixed
set codebook group number to 2 and total size to 512 (ablation)TURN codebook · VGGSScIoU and AUCImproved
Self-Cross reconstruction instead of other combinations or none (ablation)TURN · VGGSScIoU and AUCImproved

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