File size: 1,643 Bytes
c4b9a65
4836070
 
 
 
 
c4b9a65
4836070
 
 
c4b9a65
4836070
 
 
 
c4b9a65
4836070
 
 
c4b9a65
 
4836070
c4b9a65
4836070
 
 
 
 
 
c4b9a65
4836070
c4b9a65
4836070
c4b9a65
 
 
 
 
 
 
 
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
<p align="center">
  <h1 align="center"> ✏️ Data for VidChain Excercise</h1>
  <h2 align="center">VidChain: Chain-of-Tasks with Metric-based Direct Preference Optimization for Dense Video Captioning</h2>
  
<p align="center">Ji Soo Lee*, Jongha Kim*, Jeehye Na, Jinyoung Park, Hyunwoo J. Kim†.
  </p>

  <h2 align="center">
    AAAI 2025 
  </h2>

  <h3 align="center">
    <a href="https://arxiv.org/pdf/2501.06761" target='_blank'><img src="https://img.shields.io/badge/arXiv-2501.06761-b31b1b.svg"></a>
    <a href="https://huggingface.co/datasets/simplecloud/VidChain-Data"><img src="https://img.shields.io/badge/huggingface-datasets-yellow"></a>
  </h3>

<div align="center">
  <img src="asset/main.png" width="750px" />
</div>


## 🎯 Learning Objectives
By working through this exercise, you will:
- Reproduce baseline behavior of a video-language model (**VTimeLLM**, CVPR 2024 Highlight).  
- Observe the limitations of existing approaches in temporal reasoning and coherence.  
- Implement and experiment with **VidChain's improvements** using M-DPO.  
- Run inference on videos to generate **dense temporal captions (Dense Video Captioning)**.  
- Evaluate how preference alignment improves performance over baselines.  
- Discuss potential strategies for ensembling different reasoning paths of VidChain's CoTasks.  

<br>

## Citations 🌱
```
@inproceedings{lee2025vidchain,
  title={VidChain: Chain-of-Tasks with Metric-based Direct Preference Optimization for Dense Video Captioning},
  author={Lee, Ji Soo and Kim, Jongha and Na, Jeehye and Park, Jinyoung and Kim, Hyunwoo J},
  booktitle={AAAI},
  year={2025}
}
```