File size: 3,508 Bytes
7853351 | 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 43 44 45 46 47 48 49 50 51 52 53 54 55 56 57 58 59 60 61 62 63 64 65 66 67 68 69 70 71 72 73 74 75 76 77 78 79 80 81 82 83 84 | # Vis-Edit Instruct Dataset
**Vis-Edit Instruct** is a custom-built benchmark dataset designed to evaluate the performance of image editing models in a variety of real-world scenarios. Despite its small size, this dataset is diverse and curated to represent the complex and practical use cases encountered in image editing tasks.
## 📦 Dataset Overview
- **Total Images**: 35
- **Total Editing Requests**: 90
- **Main Purpose**: Benchmarking and evaluating the capabilities of image editing models under diverse and challenging conditions.
## 🖼️ Image Categories
The dataset includes 35 carefully selected images sourced from various online platforms. These images are distributed across five content categories to ensure a rich variety of contexts:
1. **People**
- Includes full-body and portrait shots
- Captures real-life activities such as playing sports, riding motorcycles, and posing
- Helps assess the model's understanding of human features and body language
2. **Natural Landscapes**
- Features forests, beaches, oceans, and rivers
- Captured under different lighting conditions and textures
- Tests object recognition and scene composition skills
3. **Animals**
- Contains dogs, cats, rabbits, elephants, and more
- Includes various poses such as sitting and running
- Evaluates model performance in animal-related edits
4. **Flowers and Plants**
- Diverse flora types with intricate textures and vibrant colors
- Designed to test the model's fine-grained recognition ability
5. **Urban Contexts**
- Scenes include fashion boutiques, cafes, residential buildings, and street views
- Designed to assess the model’s capability in handling complex urban environments
Each category is carefully balanced to ensure comprehensive representation of real-world scenarios.
## 🧪 Editing Request Scenarios
To assess editing models effectively, **Vis-Edit Instruct** includes 90 editing instructions categorized into six key scenarios (15 instructions per scenario):
1. **Sufficient Information**
- Requests provide clear and complete details for straightforward execution
2. **Lack of Interactive Object**
- The object to be modified is missing or not clearly identified
3. **Ambiguous Interactive Object**
- The targeted object appears multiple times, creating uncertainty
4. **Unsupported Tasks**
- Instructions include tasks that go beyond current model capabilities
5. **Insufficient Specific Parameters**
- Lacks precise instructions, emphasizing the need for detailed user input
6. **Parameter Suggestions**
- Requests come with user-suggested parameters, testing compliance and precision
These scenarios reflect real user interactions and serve as valuable challenges for evaluating and improving image editing systems.
## 📊 Category Distribution
A distribution chart is provided in the dataset (`category_distribution.png`) to visualize the representation of each category.
## 🔍 Applications
This dataset is ideal for:
- Evaluating multimodal image editing models
- Benchmarking instruction-following capabilities
- Testing robustness across ambiguous or underspecified prompts
- Enhancing adaptability in real-world image editing scenarios
## 📜 License
This dataset is released for academic research purposes. Please ensure proper citation when used.
---
For more details, example images, and baseline results, please refer to the documentation and sample folder provided in this repository.
|