AI Paper Atlas

Research map / Vision-language and multimodal / CLIP and zero-shot

EventBind: Learning a Unified Representation to Bind Them All for Event-based Open-world Understanding

jiazhou zhou, Xu Zheng, Yuanhuiyi Lyu, Lin Wang

ECCV 2024 · Vision-language and multimodal › CLIP and zero-shot · approach: Recaptioning and text-supervised pretraining · 40 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 HTCA module (ablation)EventBind · N-Caltech101top-1 accuracy (%): 50.40 -> 92.54Improved
add hybrid text prompts (ablation)EventBind · N-Caltech101top-1 accuracy (%): 92.54 -> 93.40Improved
add proposed event encoder (ablation)EventBind · N-Caltech101top-1 accuracy (%): 93.40 -> 94.08Improved
add event prompts and cross-frame temporal modeling (ablation)event encoder · N-Caltech101top-1 accuracy (%): 90.40 -> 94.08Improved
add semantic consistency loss when image input available (ablation)HTCA · N-Caltech101top-1 accuracy (%): 93.11 -> 94.08Improved
fine-tune CLIP image encoder rather than freezing it (ablation)event encoder · N-Caltech101top-1 accuracy (%): 86.69 -> 94.08Improved
use RGB event frames instead of gray-scale (ablation)event representation · N-Caltech101top-1 accuracy (%): 92.18 -> 94.08Improved
add image and event content prompts (ablation)text encoder · N-Caltech101top-1 accuracy (%): 91.96 -> 94.08Improved
increase event prompt length from 4 to 16 (ablation)event encoder · N-Caltech101top-1 accuracy (%): 93.27 -> 94.08Improved
use EventBind (ViT-L-14) on N-ImageNet fine-tuning vs smaller backbonesEventBind · N-ImageNet fine-tuningtop-1 accuracy (%): 42.94 (ViT-B-32) -> 63.54 (ViT-L-14)Improved
use EventBind ViT-B-16 instead of EventCLIP ViT-L-14 on N-Caltech101 for efficiencyEventBind · N-Caltech101top-1 accuracy (%): 93.57 -> 94.08 with 635MB vs 890MBImproved

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