website

princetonvisualai · Jan 9, 2025 · 96d66d7 · 96d66d7
1 parent 29b39e6
commit 96d66d7
Show file tree

Hide file tree

Showing 2 changed files with 48 additions and 8 deletions.
diff --git a/README.md b/README.md
@@ -1,8 +1,48 @@
-# icons 
+# ICONS
 
-ICONS: Influence Consensus for Vision-Language Data Selection
+**ICONS: Influence Consensus for Vision-Language Data Selection**
 
-Xindi Wu, Mengzhou Xia, Rulin Shao, Zhiwei Deng, Pang Wei Koh, Olga Russakovsky
+[[paper](https://arxiv.org/abs/2501.00654)][[website](https://princetonvisualai.github.io/icons/)][[dataset](https://huggingface.co/datasets/xindiw/LLAVA-ICONS-133K)]
+
+Authors: [Xindi Wu](https://xindiwu.github.io/), [Mengzhou Xia](https://xiamengzhou.github.io/), [Rulin Shao](https://rulinshao.github.io/), [Zhiwei Deng](https://lucas2012.github.io/), [Pang Wei Koh](https://koh.pw/), [Olga Russakovsky](https://www.cs.princeton.edu/~olgarus/)
+
+We propose ICONS, a method for selecting vision-language data that optimizes training efficiency by identifying and prioritizing data samples that are consistently valuable across multiple tasks.
+
+## News 🔥
+- [01/25] We have released the LLAVA-ICONS-133K dataset on [Hugging Face](https://huggingface.co/datasets/xindiw/LLAVA-ICONS-133K) for public use.
+- [12/24] We have released the paper [ICONS](https://arxiv.org/abs/2501.00654).
+
+## Table of Contents
+
+- [Installation](#installation)
+- [Usage](#usage)
+- [Configuration](#configuration)
+- [Contributing](#contributing)
+- [License](#license)
+
+## Installation
+
+To set up the environment for ICONS, you can use the provided `environment.yml` file to create a Conda environment.
+
+```bash
+conda env create -f environment.yml
+conda activate icons
+```
+
+## Usage
+Stage 1: Specialist 
+
+Stage 2: Generalist
+
+## Citation
+If you find this repository useful for your research, please cite with the following BibTeX entry:
+```
+@article{wu2024icons,
+  title={ICONS: Influence Consensus for Vision-Language Data Selection},
+  author={Wu, Xindi and Xia, Mengzhou and Shao, Rulin and Deng, Zhiwei and Koh, Pang Wei and Russakovsky, Olga},
+  journal={arXiv preprint arXiv:2501.00654},
+  year={2024}
+}
+```
 
 
-Under construction 🚧
diff --git a/docs/index.html b/docs/index.html
@@ -236,10 +236,9 @@ <h1 class="title">ICONS: Influence Consensus for Vision-Language Data Selection<
         <a href="https://github.com/princetonvisualai/icons" class="button">
           <i class="fab fa-github"></i> Code
         </a>
-        <a href="https://huggingface.co/datasets/xdiwu/LLaVA-ICONS-133K" class="button">
+        <a href="https://huggingface.co/datasets/xindiw/LLAVA-ICONS-133K" class="button">
           <span class="huggingface-small">🤗</span> Dataset
         </a>
-        <p style="margin-top: 15px; color: #666;">Website under construction 🚧</p>
       </div>
     </div>
   </div>
@@ -746,7 +745,7 @@ <h3>Ablation Studies</h3>
             <!-- Different Aggregation Approaches -->
             <div style="flex: 1; padding: 20px; background: #f8f9fa; border-radius: 10px; overflow: hidden;">
               <h4>Different Aggregation Approaches</h4>
-              <p>To understand how different methods of combining task-specific influence scores affect the performance, we compare our majority voting approach (<strong>Vote</strong>) with four aggregation approaches: (1) <strong>Mean</strong>, (2) <strong>Max</strong>, (3) <strong>Rank</strong>, (4) <strong>Norm</strong>.</p>
+              <p>To understand how different methods of combining task-specific influence scores affect the performance, we compare our majority voting approach (<strong>Vote</strong>) with four aggregation methods: (1) <strong>Mean</strong>, (2) <strong>Max</strong>, (3) <strong>Rank</strong>, (4) <strong>Norm</strong>.</p>
               <table style="width: 100%; border-collapse: collapse;">
                 <thead>
                   <tr style="background-color: #e9ecef;">
@@ -855,7 +854,8 @@ <h4>Different Aggregation Approaches</h4>
             <!-- Consensus-aware Selection vs. Direct-merge Selection -->
             <div style="flex: 1; padding: 20px; background: #f8f9fa; border-radius: 10px; overflow: hidden;">
               <h4>Consensus-aware Selection vs. Direct-merge Selection</h4>
-              <p>To validate our choice of a consensus-aware approach versus a direct-merge approach, we compare the performance of both approaches.</p>
+              <p>To validate our choice of a consensus-aware approach versus a direct-merge approach (directly
+                computing influence using a combined validation set), we compare the performance of both approaches.</p>
               <table style="width: 100%; border-collapse: collapse;">
                 <thead>
                   <tr style="background-color: #e9ecef;">