<?xml version="1.0" encoding="utf-8"?><feed xmlns="http://www.w3.org/2005/Atom" xml:lang="ko"><generator uri="https://jekyllrb.com/" version="4.4.1">Jekyll</generator><link href="https://limhapark.github.io/feed.xml" rel="self" type="application/atom+xml" /><link href="https://limhapark.github.io/" rel="alternate" type="text/html" hreflang="ko" /><updated>2026-09-01T14:06:13+09:00</updated><id>https://limhapark.github.io/feed.xml</id><title type="html">Halim Park, 박하림</title><subtitle></subtitle><author><name>박하림</name></author><entry><title type="html">NeRF: Representing Scenes as Neural Radiance Fields for View Synthesis</title><link href="https://limhapark.github.io/posts/nerf/" rel="alternate" type="text/html" title="NeRF: Representing Scenes as Neural Radiance Fields for View Synthesis" /><published>2026-08-30T00:00:00+09:00</published><updated>2026-08-30T00:00:00+09:00</updated><id>https://limhapark.github.io/posts/nerf</id><content type="html" xml:base="https://limhapark.github.io/posts/nerf/"><![CDATA[<blockquote>
  <p><strong>한 줄 요약</strong> — 5차원 입력 $(x, y, z, \theta, \phi)$ 를 받아 색과 밀도를 내보내는 MLP와 볼륨 렌더링을 결합해 사실적인 새 시점 이미지를 만든다.</p>
</blockquote>

<p><strong>논문</strong>: <a href="https://arxiv.org/abs/2003.08934">arXiv:2003.08934</a></p>

<h2 id="1-문제-정의">1. 문제 정의</h2>

<h2 id="2-방법">2. 방법</h2>

<h2 id="3-실험">3. 실험</h2>

<h2 id="4-정리">4. 정리</h2>]]></content><author><name>박하림</name></author><category term="논문리뷰" /><category term="3D Vision" /><category term="NeRF" /><category term="Novel View Synthesis" /><category term="Implicit Representation" /><summary type="html"><![CDATA[장면을 좌표 → 색/밀도 함수로 표현하는 MLP를 학습해 새로운 시점의 이미지를 합성한다.]]></summary></entry><entry><title type="html">An Image is Worth 16x16 Words: Transformers for Image Recognition at Scale (ViT)</title><link href="https://limhapark.github.io/posts/vit/" rel="alternate" type="text/html" title="An Image is Worth 16x16 Words: Transformers for Image Recognition at Scale (ViT)" /><published>2026-08-28T00:00:00+09:00</published><updated>2026-08-28T00:00:00+09:00</updated><id>https://limhapark.github.io/posts/vit</id><content type="html" xml:base="https://limhapark.github.io/posts/vit/"><![CDATA[<blockquote>
  <p><strong>한 줄 요약</strong> — 이미지를 16×16 패치 시퀀스로 만들어 표준 Transformer에 넣으면, 대규모 사전학습 조건에서 CNN보다 좋은 성능을 낸다.</p>
</blockquote>

<p><strong>논문</strong>: <a href="https://arxiv.org/abs/2010.11929">arXiv:2010.11929</a></p>

<h2 id="1-문제-정의">1. 문제 정의</h2>

<h2 id="2-방법">2. 방법</h2>

<h2 id="3-실험">3. 실험</h2>

<h2 id="4-정리">4. 정리</h2>]]></content><author><name>박하림</name></author><category term="논문리뷰" /><category term="Computer Vision" /><category term="Transformer" /><category term="Image Classification" /><category term="Pretraining" /><summary type="html"><![CDATA[이미지를 패치 단위로 잘라 토큰처럼 다루면 순수 Transformer만으로도 CNN을 능가한다.]]></summary></entry><entry><title type="html">Denoising Diffusion Probabilistic Models (DDPM)</title><link href="https://limhapark.github.io/posts/ddpm/" rel="alternate" type="text/html" title="Denoising Diffusion Probabilistic Models (DDPM)" /><published>2026-08-25T00:00:00+09:00</published><updated>2026-08-25T00:00:00+09:00</updated><id>https://limhapark.github.io/posts/ddpm</id><content type="html" xml:base="https://limhapark.github.io/posts/ddpm/"><![CDATA[<blockquote>
  <p><strong>한 줄 요약</strong> — 데이터를 점진적으로 가우시안 노이즈로 망가뜨리는 과정을 정의하고, 그 역과정을 신경망으로 학습해 샘플을 생성한다.</p>
</blockquote>

<p><strong>논문</strong>: <a href="https://arxiv.org/abs/2006.11239">arXiv:2006.11239</a> · <strong>코드</strong>: <a href="https://github.com/hojonathanho/diffusion">GitHub</a></p>

<h2 id="1-문제-정의">1. 문제 정의</h2>

<!-- 이 논문이 풀려는 문제와 기존 방법의 한계를 적습니다. -->

<h2 id="2-방법">2. 방법</h2>

<p>Forward process는 다음과 같이 정의됩니다.</p>

\[q(\mathbf{x}_t \mid \mathbf{x}_{t-1}) = \mathcal{N}(\mathbf{x}_t; \sqrt{1-\beta_t}\,\mathbf{x}_{t-1}, \beta_t \mathbf{I})\]

<h2 id="3-실험">3. 실험</h2>

<table>
  <thead>
    <tr>
      <th>Dataset</th>
      <th>FID ↓</th>
      <th>IS ↑</th>
    </tr>
  </thead>
  <tbody>
    <tr>
      <td>CIFAR-10</td>
      <td>3.17</td>
      <td>9.46</td>
    </tr>
  </tbody>
</table>

<h2 id="4-정리">4. 정리</h2>

<!-- 내 생각, 한계, 후속 연구와의 관계 -->]]></content><author><name>박하림</name></author><category term="논문리뷰" /><category term="Generative Models" /><category term="Diffusion" /><category term="Image Generation" /><summary type="html"><![CDATA[노이즈를 조금씩 제거하는 과정을 학습해 이미지를 생성하는 확산 모델의 기본 형태.]]></summary></entry></feed>