Update README.md
Browse files
README.md
CHANGED
|
@@ -33,21 +33,15 @@ configs:
|
|
| 33 |
This dataset accompanies the paper [TabComp: A Dataset for Visual Table Reading Comprehension](https://aclanthology.org/2025.findings-naacl.320.pdf)
|
| 34 |
|
| 35 |
TabComp evaluates **Vision-Language Models (VLMs)** on their ability to **read, understand, and reason over table images** without relying on OCR, using **generative question answering**.
|
| 36 |
-
|
| 37 |
---
|
| 38 |
-
|
| 39 |
## 🔍 Why TabComp?
|
| 40 |
-
|
| 41 |
Modern VLMs perform well on general VQA but struggle with **tables**, which require:
|
| 42 |
-
|
| 43 |
- Structured reasoning across rows/columns
|
| 44 |
- Understanding layout + text jointly
|
| 45 |
- Multi-step inference over semi-structured data
|
| 46 |
|
| 47 |
👉 TabComp isolates this challenge and provides a **focused benchmark for table understanding**.
|
| 48 |
-
|
| 49 |
---
|
| 50 |
-
|
| 51 |
## 📊 Dataset Overview
|
| 52 |
|
| 53 |
- **Images:** 3,318 table images
|
|
@@ -64,18 +58,14 @@ Given:
|
|
| 64 |
|
| 65 |
Generate:
|
| 66 |
- A **natural language answer** requiring table comprehension
|
| 67 |
-
|
| 68 |
---
|
| 69 |
-
|
| 70 |
## 🧠 What Makes It Challenging?
|
| 71 |
|
| 72 |
- ❌ No OCR signals
|
| 73 |
- ✅ Dense textual + structural information
|
| 74 |
- ✅ Long-range dependencies across table cells
|
| 75 |
- ✅ Generative answers (not extractive spans)
|
| 76 |
-
|
| 77 |
---
|
| 78 |
-
|
| 79 |
## 📁 Data Format
|
| 80 |
|
| 81 |
🏆 Leaderboard (Baseline Results)
|
|
|
|
| 33 |
This dataset accompanies the paper [TabComp: A Dataset for Visual Table Reading Comprehension](https://aclanthology.org/2025.findings-naacl.320.pdf)
|
| 34 |
|
| 35 |
TabComp evaluates **Vision-Language Models (VLMs)** on their ability to **read, understand, and reason over table images** without relying on OCR, using **generative question answering**.
|
|
|
|
| 36 |
---
|
|
|
|
| 37 |
## 🔍 Why TabComp?
|
|
|
|
| 38 |
Modern VLMs perform well on general VQA but struggle with **tables**, which require:
|
|
|
|
| 39 |
- Structured reasoning across rows/columns
|
| 40 |
- Understanding layout + text jointly
|
| 41 |
- Multi-step inference over semi-structured data
|
| 42 |
|
| 43 |
👉 TabComp isolates this challenge and provides a **focused benchmark for table understanding**.
|
|
|
|
| 44 |
---
|
|
|
|
| 45 |
## 📊 Dataset Overview
|
| 46 |
|
| 47 |
- **Images:** 3,318 table images
|
|
|
|
| 58 |
|
| 59 |
Generate:
|
| 60 |
- A **natural language answer** requiring table comprehension
|
|
|
|
| 61 |
---
|
|
|
|
| 62 |
## 🧠 What Makes It Challenging?
|
| 63 |
|
| 64 |
- ❌ No OCR signals
|
| 65 |
- ✅ Dense textual + structural information
|
| 66 |
- ✅ Long-range dependencies across table cells
|
| 67 |
- ✅ Generative answers (not extractive spans)
|
|
|
|
| 68 |
---
|
|
|
|
| 69 |
## 📁 Data Format
|
| 70 |
|
| 71 |
🏆 Leaderboard (Baseline Results)
|