initial publishing

Pseudo-Lab · Jan 10, 2025 · 0542fef · 0542fef
1 parent f025c8a
commit 0542fef
Show file tree

Hide file tree

Showing 11 changed files with 36 additions and 42 deletions.
diff --git a/_images/Untitled.png b/_images/Untitled.png
diff --git a/_sources/docs/review/zero123.md b/_sources/docs/review/zero123.md
@@ -1,4 +1,4 @@
-# zero 1-to-3 : Zero-shot One Image to 3D Object
+# zero 123
 
 ``` {admonition} Information
 - **Title:** zero 1-to-3 : Zero-shot One Image to 3D Object
@@ -46,13 +46,9 @@
 4. **Zero 1-to-3 를 제안** 
     - **Zero 1-to-3 :**
         - Stable Diffusion과 같은 large-scale diffusion 모델이 단일 RGB 이미지를 입력으로,
-
-        <aside>
         1. **zero-shot novel view synthesis**
         2. **3D shape reconstruction**
-        </aside>
-        
-              를 수행하기 위해 **카메라 view point(시점)을 조작을 컨트롤하는 매커니즘 학습**
+        를 수행하기 위해 **카메라 view point(시점)을 조작을 컨트롤하는 매커니즘 학습**
 
     - 주어진 단일 RGB 이미지 입력하에 이러한 TASK를 수행 하는 것은 매우 많은 제약이 존재 
     → Diffusion 모델을 기반으로 다양한 시점에서 방대한 객체 이미지를 생성하여 이용
@@ -86,7 +82,7 @@
     - [**기존 접근 방식**:](https://velog.io/@dldydldy75/3D-Understanding)
 
         :::{figure-md} 
-        <img src="../../pics/zero123/image 1.png" alt="tag" class="bg-primary mb-1">
+        <img src="../../pics/zero123/image1.png" alt="tag" class="bg-primary mb-1">
         :::
 
         1. **데이터 수집 형태에 따른 전역적인(global) 특성 기반의 조건화 모델** 
@@ -134,7 +130,7 @@
     2. **인터넷 규모 데이터셋에 반영된 뷰포인트 편향** : 생성 모델들은 인터넷에서 반영된 시점 편향을 물려받아, 특정한 자세 및 시점에서의 이미지를 생성하는 경향 존재
 
     :::{figure-md} 
-    <img src="../../pics/zero123/image 2.png" alt="tag" class="bg-primary mb-1">
+    <img src="../../pics/zero123/image2.png" alt="tag" class="bg-primary mb-1">
     :::
 
 
@@ -146,14 +142,14 @@
 - **접근법 [Figure3]**
 
     :::{figure-md} 
-    <img src="../../pics/zero123/image 3.png" alt="tag" class="bg-primary mb-1">
+    <img src="../../pics/zero123/image3.png" alt="tag" class="bg-primary mb-1">
     :::
 
 - 사전 훈련된 diffusion 모델을 미세조정하여 나머지 표현을 손상시키지 않고 카메라 파라미터를 제어하도록 학습.
 - **Latent Diffusion Architecture 를 이용**
     - **[참고] LDM**
     :::{figure-md} 
-    <img src="../../pics/zero123/image 4.png" alt="tag" class="bg-primary mb-1">
+    <img src="../../pics/zero123/image4.png" alt="tag" class="bg-primary mb-1">
     :::
     - Encoder($\mathcal{E}$), Denoiser(U-Net, $\epsilon_\theta$), Decoder($\mathcal{D}$)로 구성
     - **The Objective**
@@ -212,7 +208,7 @@
     - $\nabla \mathcal{L}_{SJC}$ : PAAS 스코어
 
     :::{figure-md} 
-    <img src="../../pics/zero123/image 7.png" alt="tag" class="bg-primary mb-1">
+    <img src="../../pics/zero123/image7.png" alt="tag" class="bg-primary mb-1">
     :::
 
 - 입력 뷰와의 **MSE** 손실로 최적화
@@ -312,18 +308,18 @@
 ## 4.4. Novel View Synthesis Results
 
 :::{figure-md} 
-<img src="../../pics/zero123/image 12.png" alt="tag" class="bg-primary mb-1">
+<img src="../../pics/zero123/image12.png" alt="tag" class="bg-primary mb-1">
 :::
 - **Table 2** : RTMV에서의 새로운 뷰 합성 결과.  RTMV의 장면은 Objaverse 훈련 데이터와 분포가 다르지만, 우리의 모델은 여전히 기준선을 상당한 차이로 능가.
 
 :::{figure-md} 
-<img src="../../pics/zero123/image 13.png" alt="tag" class="bg-primary mb-1">
+<img src="../../pics/zero123/image13.png" alt="tag" class="bg-primary mb-1">
 :::
 - **Figure 5 : Novel view synthesis on Google Scanned Object** 
 왼쪽에 표시된 입력 뷰는 두 개의 무작위 샘플링된 새로운 뷰를 합성하는 데 사용됨. 해당하는 실제 뷰는 오른쪽에 표시되어 있음. 기준 방법들과 비교할 때, 우리가 합성한 새로운 뷰는 실제와 매우 일치하는 풍부한 텍스트 및 기하학적 세부 사항을 포함하고 있으며, 반면 기준 방법들은 고주파 세부 사항의 유의미한 손실을 보임.
 
 :::{figure-md} 
-<img src="../../pics/zero123/image 14.png" alt="tag" class="bg-primary mb-1">
+<img src="../../pics/zero123/image14.png" alt="tag" class="bg-primary mb-1">
 :::
 - **Figure 6 Novel view synthesis on RTMV**
 왼쪽에 표시된 입력 뷰는 두 개의 무작위로 샘플링된 새로운 뷰를 합성하는 데 사용됨. 오른쪽에는 해당하는 실제 뷰가 표시됨. 우리가 합성한 뷰는 큰 카메라 시점 변화가 있을 때조차도 높은 충실도를 유지하며, 대부분의 다른 방법들은 품질이 급격히 저하됨.
@@ -333,7 +329,7 @@
     - 그러나 생성된 포인트 클라우드의 크기가 작아 Point-E가 새로운 뷰 합성(novel view synthesis)에서의 적용 가능성을 제한함.
 
 :::{figure-md} 
-<img src="../../pics/zero123/image 15.png" alt="tag" class="bg-primary mb-1">
+<img src="../../pics/zero123/image15.png" alt="tag" class="bg-primary mb-1">
 :::
 
 - **Figure 7 Novel view synthesis on in-the-wild images.**
@@ -343,7 +339,7 @@
     - 입력 이미지가 2D이기 때문에 항상 객체의 부분적인 뷰만을 나타내고 많은 부분이 관찰되지 않으므로, diffusion 을 통해 다양한 시점에서 샘플들을 랜덤으로 생성
 
 :::{figure-md} 
-<img src="../../pics/zero123/image 16.png" alt="tag" class="bg-primary mb-1">
+<img src="../../pics/zero123/image16.png" alt="tag" class="bg-primary mb-1">
 :::
 
 - Figure 8: 입력 뷰를 고정하고, 다양한 시점에서 새로운 샘플을 랜덤으로 생성하여 시각화
@@ -352,16 +348,16 @@
 ## 4.5. 3D Reconstruction Results
 
 :::{figure-md} 
-<img src="../../pics/zero123/image 17.png" alt="tag" class="bg-primary mb-1">
+<img src="../../pics/zero123/image17.png" alt="tag" class="bg-primary mb-1">
 :::
 
 - 실제 ground truth 와 유사한 고충실도의 3D 메쉬를 reconstruct
 
 :::{figure-md} 
-<img src="../../pics/zero123/image 18.png" alt="tag" class="bg-primary mb-1">
+<img src="../../pics/zero123/image18.png" alt="tag" class="bg-primary mb-1">
 :::
 :::{figure-md} 
-<img src="../../pics/zero123/image 19.png" alt="tag" class="bg-primary mb-1">
+<img src="../../pics/zero123/image19.png" alt="tag" class="bg-primary mb-1">
 :::
 
 - **MCC (Multiview Compressive Coding)**:
@@ -376,7 +372,7 @@
 ## 4.6. Text to Image to 3D-
 
 :::{figure-md} 
-<img src="../../pics/zero123/image 20.png" alt="tag" class="bg-primary mb-1">
+<img src="../../pics/zero123/image20.png" alt="tag" class="bg-primary mb-1">
 :::
 
 - 실제 환경에서 촬영된 이미지 외에도, Dall-E-2와 같은 txt2img 모델이 생성한 이미지에 대해서도 테스트

diff --git a/_sources/intro.md b/_sources/intro.md
@@ -45,4 +45,4 @@ CVPR 2024 Workshop
 - 한동현 [Donghyun Han] | [GitHub](https://github.com/donghyun99) | [Linkedin](https://www.linkedin.com/in/donghyun99/) |
 - 이창환 [ChangHwan Lee] | [Github](https://github.com/Hwan-I) | Linkedin 
 - 유경민 [Kyeongmin Yu] | [Github](https://github.com/yukyeongmin) | Linkdedin |
-- 이정인 [Jeongin Lee] | Github | Linkdedin |
+- 이정인 [Jeongin Lee] | Github | [Linkdedin](https://www.linkedin.com/in/innimu/) |
diff --git a/docs/review/DreamFusion.html b/docs/review/DreamFusion.html
@@ -321,7 +321,7 @@
 
 
 
-<li class="toctree-l1"><a class="reference internal" href="zero123.html">zero 1-to-3 : Zero-shot One Image to 3D Object</a></li>
+<li class="toctree-l1"><a class="reference internal" href="zero123.html">zero 123</a></li>
 
 
 

diff --git a/docs/review/DreamPose.html b/docs/review/DreamPose.html
@@ -321,7 +321,7 @@
 
 
 
-<li class="toctree-l1"><a class="reference internal" href="zero123.html">zero 1-to-3 : Zero-shot One Image to 3D Object</a></li>
+<li class="toctree-l1"><a class="reference internal" href="zero123.html">zero 123</a></li>
 
 
 

diff --git a/docs/review/zero123.html b/docs/review/zero123.html
@@ -9,7 +9,7 @@
     <meta charset="utf-8" />
     <meta name="viewport" content="width=device-width, initial-scale=1.0" /><meta name="generator" content="Docutils 0.18.1: http://docutils.sourceforge.net/" />
 
-    <title>zero 1-to-3 : Zero-shot One Image to 3D Object &#8212; Text-to-Image Generation-feat-Diffusion</title>
+    <title>zero 123 &#8212; Text-to-Image Generation-feat-Diffusion</title>
 
 
 
@@ -321,7 +321,7 @@
 
 
 
-<li class="toctree-l1 current active"><a class="current reference internal" href="#">zero 1-to-3 : Zero-shot One Image to 3D Object</a></li>
+<li class="toctree-l1 current active"><a class="current reference internal" href="#">zero 123</a></li>
 
 
 
@@ -534,7 +534,7 @@
 
 
 <div id="jb-print-docs-body" class="onlyprint">
-    <h1>zero 1-to-3 : Zero-shot One Image to 3D Object</h1>
+    <h1>zero 123</h1>
     <!-- Table of contents -->
     <div id="print-main-content">
         <div id="jb-print-toc">
@@ -544,7 +544,7 @@ <h2> Contents </h2>
             </div>
             <nav aria-label="Page">
                 <ul class="visible nav section-nav flex-column">
-<li class="toc-h1 nav-item toc-entry"><a class="reference internal nav-link" href="#">zero 1-to-3 : Zero-shot One Image to 3D Object</a></li>
+<li class="toc-h1 nav-item toc-entry"><a class="reference internal nav-link" href="#">zero 123</a></li>
 <li class="toc-h1 nav-item toc-entry"><a class="reference internal nav-link" href="#abstract-zero-shot-one-image-to-3d-object">Abstract : Zero-shot One Image to 3D Object</a></li>
 <li class="toc-h1 nav-item toc-entry"><a class="reference internal nav-link" href="#introduction">1. Introduction</a></li>
 <li class="toc-h1 nav-item toc-entry"><a class="reference internal nav-link" href="#related-work">2. Related Work</a></li>
@@ -590,8 +590,8 @@ <h2> Contents </h2>
 <div id="searchbox"></div>
                 <article class="bd-article">
 
-  <section class="tex2jax_ignore mathjax_ignore" id="zero-1-to-3-zero-shot-one-image-to-3d-object">
-<h1>zero 1-to-3 : Zero-shot One Image to 3D Object<a class="headerlink" href="#zero-1-to-3-zero-shot-one-image-to-3d-object" title="Permalink to this heading">#</a></h1>
+  <section class="tex2jax_ignore mathjax_ignore" id="zero-123">
+<h1>zero 123<a class="headerlink" href="#zero-123" title="Permalink to this heading">#</a></h1>
 <div class="admonition-information admonition">
 <p class="admonition-title">Information</p>
 <ul class="simple">
@@ -656,13 +656,11 @@ <h1>1. Introduction<a class="headerlink" href="#introduction" title="Permalink t
 <ul class="simple">
 <li><p>Stable Diffusion과 같은 large-scale diffusion 모델이 단일 RGB 이미지를 입력으로,</p></li>
 </ul>
-  <aside>
-  1. **zero-shot novel view synthesis**
-  2. **3D shape reconstruction**
-  </aside>
-<div class="highlight-none notranslate"><div class="highlight"><pre><span></span>    를 수행하기 위해 **카메라 view point(시점)을 조작을 컨트롤하는 매커니즘 학습**
-</pre></div>
-</div>
+<ol class="arabic simple">
+<li><p><strong>zero-shot novel view synthesis</strong></p></li>
+<li><p><strong>3D shape reconstruction</strong>
+를 수행하기 위해 <strong>카메라 view point(시점)을 조작을 컨트롤하는 매커니즘 학습</strong></p></li>
+</ol>
 </li>
 <li><p>주어진 단일 RGB 이미지 입력하에 이러한 TASK를 수행 하는 것은 매우 많은 제약이 존재
 → Diffusion 모델을 기반으로 다양한 시점에서 방대한 객체 이미지를 생성하여 이용</p>
@@ -1247,7 +1245,7 @@ <h2>D. 3D Reconstruction<a class="headerlink" href="#d-3d-reconstruction" title=
   </div>
   <nav class="bd-toc-nav page-toc">
     <ul class="visible nav section-nav flex-column">
-<li class="toc-h1 nav-item toc-entry"><a class="reference internal nav-link" href="#">zero 1-to-3 : Zero-shot One Image to 3D Object</a></li>
+<li class="toc-h1 nav-item toc-entry"><a class="reference internal nav-link" href="#">zero 123</a></li>
 <li class="toc-h1 nav-item toc-entry"><a class="reference internal nav-link" href="#abstract-zero-shot-one-image-to-3d-object">Abstract : Zero-shot One Image to 3D Object</a></li>
 <li class="toc-h1 nav-item toc-entry"><a class="reference internal nav-link" href="#introduction">1. Introduction</a></li>
 <li class="toc-h1 nav-item toc-entry"><a class="reference internal nav-link" href="#related-work">2. Related Work</a></li>

diff --git a/genindex.html b/genindex.html
@@ -318,7 +318,7 @@
 
 
 
-<li class="toctree-l1"><a class="reference internal" href="docs/review/zero123.html">zero 1-to-3 : Zero-shot One Image to 3D Object</a></li>
+<li class="toctree-l1"><a class="reference internal" href="docs/review/zero123.html">zero 123</a></li>
 
 
 

diff --git a/intro.html b/intro.html
@@ -320,7 +320,7 @@
 
 
 
-<li class="toctree-l1"><a class="reference internal" href="docs/review/zero123.html">zero 1-to-3 : Zero-shot One Image to 3D Object</a></li>
+<li class="toctree-l1"><a class="reference internal" href="docs/review/zero123.html">zero 123</a></li>
 
 
 
@@ -604,7 +604,7 @@ <h2>Contributors<a class="headerlink" href="#contributors" title="Permalink to t
 <li><p>한동현 [Donghyun Han] | <a class="reference external" href="https://github.com/donghyun99">GitHub</a> | <a class="reference external" href="https://www.linkedin.com/in/donghyun99/">Linkedin</a> |</p></li>
 <li><p>이창환 [ChangHwan Lee] | <a class="reference external" href="https://github.com/Hwan-I">Github</a> | Linkedin</p></li>
 <li><p>유경민 [Kyeongmin Yu] | <a class="reference external" href="https://github.com/yukyeongmin">Github</a> | Linkdedin |</p></li>
-<li><p>이정인 [Jeongin Lee] | Github | Linkdedin |</p></li>
+<li><p>이정인 [Jeongin Lee] | Github | <a class="reference external" href="https://www.linkedin.com/in/innimu/">Linkdedin</a> |</p></li>
 </ul>
 </section>
 <div class="toctree-wrapper compound">

diff --git a/objects.inv b/objects.inv
diff --git a/search.html b/search.html
@@ -320,7 +320,7 @@
 
 
 
-<li class="toctree-l1"><a class="reference internal" href="docs/review/zero123.html">zero 1-to-3 : Zero-shot One Image to 3D Object</a></li>
+<li class="toctree-l1"><a class="reference internal" href="docs/review/zero123.html">zero 123</a></li>
 
 
 

diff --git a/searchindex.js b/searchindex.js
Original file line number	Diff line number	Diff line change
Expand Up		@@ -321,7 +321,7 @@



		<li class="toctree-l1"><a class="reference internal" href="zero123.html">zero 1-to-3 : Zero-shot One Image to 3D Object</a></li>
		<li class="toctree-l1"><a class="reference internal" href="zero123.html">zero 123</a></li>



Expand Down
Original file line number	Diff line number	Diff line change
Expand Up		@@ -318,7 +318,7 @@



		<li class="toctree-l1"><a class="reference internal" href="docs/review/zero123.html">zero 1-to-3 : Zero-shot One Image to 3D Object</a></li>
		<li class="toctree-l1"><a class="reference internal" href="docs/review/zero123.html">zero 123</a></li>



Expand Down
Original file line number	Diff line number	Diff line change
Expand Up		@@ -320,7 +320,7 @@



		<li class="toctree-l1"><a class="reference internal" href="docs/review/zero123.html">zero 1-to-3 : Zero-shot One Image to 3D Object</a></li>
		<li class="toctree-l1"><a class="reference internal" href="docs/review/zero123.html">zero 123</a></li>



Expand Down