Skip to content

Commit

Permalink
website
Browse files Browse the repository at this point in the history
  • Loading branch information
@XindiWu
XindiWu authored Dec 10, 2024
1 parent 4572820 commit 5c366ea
Showing 1 changed file with 46 additions and 11 deletions.
57 changes: 46 additions & 11 deletions docs/index.html
View file
Original file line number Diff line number Diff line change
Expand Up @@ -154,6 +154,9 @@ <h2 class="subtitle is-3 publication-subtitle">A Compositional Image Generation
<div class="is-size-5 publication-authors">
<span class="author-block">Princeton University</span>
</div>
<div class="is-size-5 publication-authors">
<span class="author-block" style="font-size: 0.9em; color: #666;">* Equal Contribution</span>
</div>
<div class="is-size-5 publication-authors">
<span class="author-block conference-text">[NeurIPS 2024 D&B]</span>
</div>
Expand Down Expand Up @@ -206,6 +209,12 @@ <h2 class="subtitle is-3 publication-subtitle">A Compositional Image Generation
<span>Slides</span>
</a>
</span>
<span class="link-block">
<a href="static/data/poster.pdf" class="external-link button is-normal is-rounded is-dark">
<span class="icon"><i class="fas fa-file-image"></i></span>
<span>Poster</span>
</a>
</span>
<span class="link-block">
<a href="https://openreview.net/forum?id=MU2s9wwWLo" class="external-link button is-normal is-rounded is-dark">
<span class="icon"><i class="fas fa-external-link-alt"></i></span>
Expand Down Expand Up @@ -412,20 +421,23 @@ <h2 class="title is-3" id="leaderboard"><span class="fancy_text_color">Leaderboa
</div>
</section>




<section class="section">
<div class="container is-max-desktop">
<div class="columns is-centered has-text-centered">
<div class="column is-four-fifths">
<h2 class="title is-3" id="concept-diversity">LAION-5B Concept Diversity</h2>
<h2 class="title is-3" id="performance-concept-categories">Performance on Individual Concept Categories (k=1)</h2>
<div class="content has-text-justified">
<p>
The LAION-5B dataset is analyzed for concept diversity, the heatmap below showcases the frequency of these visual concepts in sampled captions.
We evaluate the performance of T2I models across different concept categories. Color and style are easier, with all models achieving high scores. Performance is lower for generating specific numbers of objects and spatial relationships, with varying results for texture and size. Overall, DALL·E 3 outperforms others in all categories.
</p>
<div class="has-text-centered">
<img src="static/images/caption.png" alt="Concept Diversity in LAION-5B Dataset">
<img src="static/images/individual_concept.png" alt="Performance Across Concept Categories" style="width: 100%; max-width: 800px;">
</div>
<p>
<b>Concept Diversity in LAION-5B Dataset.</b> Left: Heatmap of sampled captions shows colors and styles are most frequent; shapes and spatial relationships are least. Right: Most examples include 2-3 concepts.
<b>Performance Across Concept Categories.</b> We evaluate T2I models across concept categories, finding high scores for color and style but lower performance for object counts and spatial relationships. DALL·E 3 outperforms others across all categories.
</p>
</div>
</div>
Expand All @@ -438,16 +450,35 @@ <h2 class="title is-3" id="concept-diversity">LAION-5B Concept Diversity</h2>
<div class="container is-max-desktop">
<div class="columns is-centered has-text-centered">
<div class="column is-four-fifths">
<h2 class="title is-3" id="performance-concept-categories">Performance on Individual Concept Categories (k=1)</h2>
<h2 class="title is-3" id="compositional-performance">Performance of Compositional Generation (k > 1)</h2>
<div class="content has-text-justified">
<div class="has-text-centered">
<img src="./static/images/comparison.png" style="width: 95%; max-width: 700px;">
</div>
<p>
We evaluate the performance of T2I models across different concept categories. Color and style are easier, with all models achieving high scores. Performance is lower for generating specific numbers of objects and spatial relationships, with varying results for texture and size. Overall, DALL·E 3 outperforms others in all categories.
<b>ConceptMix Shows Stronger Discriminative Power:</b>
We compare five models using 3-in-1 and GPT4v scores (global prompt-level) from T2I-CompBench, and ConceptMix with varying difficulty levels (k). ConceptMix, with varying difficulty levels (k), clearly distinguishes model performance, with gaps widening as k increases.
</p>
</div>
</div>
</div>
</div>
</section>

<section class="section">
<div class="container is-max-desktop">
<div class="columns is-centered has-text-centered">
<div class="column is-four-fifths">
<h2 class="title is-3" id="qualitative-performance">Qualitative Performance of Different T2I Models</h2>
<div class="content has-text-justified">
<p>
We compare the qualitative performance of different T2I models (SD v1.4, SD v2.1, PixArt alpha, Playground v2.5, DALL·E 3) across varying levels of compositional complexity (k = 1...7). As prompts become more complex, image quality degrades. DALL·E 3 performs best, while SD v1.4 performs worst.
</p>
<div class="has-text-centered">
<img src="static/images/individual_concept.png" alt="Performance Across Concept Categories" style="width: 100%; max-width: 800px;">
<img src="static/images/qualitative.png" alt="Qualitative Performance Comparison" style="width: 100%; max-width: 800px;">
</div>
<p>
<b>Performance Across Concept Categories.</b> We evaluate T2I models across concept categories, finding high scores for color and style but lower performance for object counts and spatial relationships. DALL·E 3 outperforms others across all categories.
<b>Qualitative Comparison:</b> Visual comparison of generated images across different models and complexity levels (k), showing degrading performance with increasing prompt complexity.
</p>
</div>
</div>
Expand All @@ -456,17 +487,21 @@ <h2 class="title is-3" id="performance-concept-categories">Performance on Indivi
</section>



<section class="section">
<div class="container is-max-desktop">
<div class="columns is-centered has-text-centered">
<div class="column is-four-fifths">
<h2 class="title is-3" id="compositional-performance">Performance of Compositional Generation (k > 1)</h2>
<h2 class="title is-3" id="concept-diversity">LAION-5B Concept Diversity</h2>
<div class="content has-text-justified">
<p>
The LAION-5B dataset is analyzed for concept diversity, the heatmap below showcases the frequency of these visual concepts in sampled captions.
</p>
<div class="has-text-centered">
<img src="./static/images/comparison.png" style="width: 95%; max-width: 700px;">
<img src="static/images/caption.png" alt="Concept Diversity in LAION-5B Dataset">
</div>
<p>
<b>ConceptMix Shows Stronger Discriminative Power:</b> ConceptMix, with varying difficulty levels (k), clearly distinguishes model performance, with gaps widening as k increases.
<b>Concept Diversity in LAION-5B Dataset.</b> Left: Heatmap of sampled captions shows colors and styles are most frequent; shapes and spatial relationships are least. Right: Most examples include 2-3 concepts.
</p>
</div>
</div>
Expand Down

0 comments on commit 5c366ea

Please sign in to comment.