From ce033dc5c9e4208eb71a7746c4bfedc6abfabc2a Mon Sep 17 00:00:00 2001 From: Chanyoung Kim <44921488+kochanha@users.noreply.github.com> Date: Wed, 27 Nov 2024 22:08:27 +0900 Subject: [PATCH] updated --- cass/index.html | 51 +++++++---------------------------------- cass/static/js/index.js | 17 ++++++++------ 2 files changed, 18 insertions(+), 50 deletions(-) diff --git a/cass/index.html b/cass/index.html index f3cebc0..2755761 100644 --- a/cass/index.html +++ b/cass/index.html @@ -237,7 +237,7 @@

- @@ -269,42 +269,7 @@


- - - - +

We introduce CASS,
Distilling Spectral Graph for Object-Context Aware Open-Vocabulary Semantic Segmentation. @@ -348,12 +313,12 @@

Method


Overall Pipeline

-
- Main figure +
+ Main figure

We present CASS, object-level Context-Aware training-free open-vocabulary Semantic Segmentation model. - Our method distills the vision foundation model's (VFM) object-level contextual spectral graph into CLIP’s attention and refines query text embeddings towards object-specific semantics. + Our method distills the vision foundation model's (VFM) object-level contextual spectral graph into CLIP's attention and refines query text embeddings towards object-specific semantics.

@@ -363,14 +328,14 @@

Spectral Object-Level Context Distillat Main figure

Detailed illustration of our proposed training-free spectral object-level context distillation mechanism. - By matching the attention graphs of VFM and CLIP head-by-head to establish complementary relationships, and distilling the fundamental object-level context of the VFM graph to CLIP, we enhance CLIP’s ability to capture intra-object contextual coherence. + By matching the attention graphs of VFM and CLIP head-by-head to establish complementary relationships, and distilling the fundamental object-level context of the VFM graph to CLIP, we enhance CLIP's ability to capture intra-object contextual coherence.


Object Presence-Driven Object-Level Context

-
- Main figure +
+ Main figure

Detailed illustration of our object presence prior-guided text embedding adjustment module. The CLIP text encoder generates text embeddings for each object class, and the object presence prior is derived from both visual and text embeddings. diff --git a/cass/static/js/index.js b/cass/static/js/index.js index a1cbc85..5a0d261 100644 --- a/cass/static/js/index.js +++ b/cass/static/js/index.js @@ -120,21 +120,24 @@ $(document).ready(function() { const slider = document.getElementById('image-slider'); const items = Array.from(slider.getElementsByClassName('slider-item')); - // 무한 슬라이드를 위해 처음 아이템들을 복제하여 끝에 추가 + // Clone items for infinite scroll items.forEach(item => { const clone = item.cloneNode(true); slider.appendChild(clone); }); let position = 0; - const slideWidth = 16.666; // 각 슬라이드 아이템의 너비 퍼센트 + const slideWidth = 33.333; // 변경: 16.666에서 33.333으로 증가 function slide() { position -= slideWidth; slider.style.transform = `translateX(${position}%)`; - // 슬라이드가 끝에 도달하면 처음으로 리셋 - if (position <= -100) { + // 수정: 마지막 이미지를 완전히 보여주기 위해 계산 방식 변경 + const totalSlides = slider.children.length; + const endPosition = -(totalSlides - 3) * slideWidth; // 한 번에 3개의 슬라이드가 보이므로 + + if (position <= endPosition) { setTimeout(() => { slider.style.transition = 'none'; position = 0; @@ -146,10 +149,10 @@ $(document).ready(function() { } } - // 3초마다 슬라이드 - const slideInterval = setInterval(slide, 3000); + // Slide every 3 seconds + const slideInterval = setInterval(slide, 4000); - // 마우스 호버 시 일시정지 + // Pause on hover slider.addEventListener('mouseenter', () => clearInterval(slideInterval)); slider.addEventListener('mouseleave', () => setInterval(slide, 3000)); }