AI Paper Atlas

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

DEAL: Disentangle and Localize Concept-level Explanations for VLMs

Tang Li, Mengmeng Ma, Xi Peng

ECCV 2024 · Vision-language and multimodal › Human and embodied modeling · approach: Scene graph generation · 6 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 concept-level disentangle and localize losses to CLIP fine-tuningCLIP ViT-B/32 · Five datasets avg.prediction accuracy (%): 65.7 -> 78.8Improved
add DEAL objectiveCLIP ViT-B/32 · Five datasets avg.concept-level localizability: 0.593 -> 0.661Improved
remove disentanglement constraint (ablation)DEAL · ImageNetprediction accuracy (%): 70.8 -> 60.4Hurt
remove disentanglement constraint (ablation)DEAL · ImageNetdisentanglability: 0.397 -> 0.354Hurt
remove localization constraint (only disentangle) (ablation)DEAL · ImageNetprediction accuracy (%): 70.8 -> 62.7Hurt
remove localization constraint (ablation)DEAL · ImageNetlocalizability: 0.673 -> 0.623Hurt
full fine-tune CLIP without concept constraintsCLIP ViT-B/32 · ImageNetdisentanglability: 0.361 -> 0.323Hurt
full fine-tune CLIP without concept constraintsCLIP ViT-B/32 · ImageNetprediction accuracy (%): 63.2 -> 68.9Improved
use DEAL instead of standard full fine-tuning (cost: doubled training time)CLIP ViT-B/32 · ImageNetprediction accuracy (%): 68.9 -> 70.8Improved
apply DEAL for part localizationCLIP RN50 · CUB-PartmIoU: 6.31 -> 12.04Improved
apply DEAL to EuroSAT with ViT-B/32CLIP ViT-B/32 · EuroSATdisentanglability: 0.192 -> 0.192No clear effect

Earlier methods this paper compares with

Named baselines from earlier accepted papers in this paper's results tables: better than 1 of 1 on most shared results.

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