BLINK
The BLINK dataset offers a benchmark to assess the visual perception capabilities of multimodal language models (LLMs). It reformulates 14 classic computer vision tasks into 3,807 multiple choice questions with images and visual prompts. This benchmark highlights the current difficulties of multimodal LLMs in achieving human performance in visual perception.
3,807 examples, 797 MB, multiple choice questions associated with one or more images, parquet format
Apache 2.0
Description
The dataset BLINK includes 3,807 examples of multiple-choice questions for computer vision tasks that require fine visual perception, such as relative depth estimation, visual correspondence, forensic detection, and multi-view reasoning. Each question is accompanied by one or more images and a visual prompt for multimodal models.
What is this dataset for?
- Testing the visual perceptual abilities of multimodal language models
- Evaluate visual comprehension in complex tasks close to human perception
- Stimulate research to improve the perception of multimodal models
Can it be enriched or improved?
The dataset can be extended with other visual perception tasks or enriched with additional annotations. Integrating additional data or refining visual prompts could improve its relevance for fine-tuning.
🔎 In summary
🧠 Recommended for
- Computer vision researchers
- Multimodal model developers
- AI R&D teams
🔧 Compatible tools
- Hugging Face Datasets
- PyTorch
- TensorFlow
- Multimodal assessment tools
💡 Tip
Use the multiple nature of the images per question to test the visual robustness of the models.
Frequently Asked Questions
What tasks are covered by the BLINK dataset?
The dataset covers 14 classic computer vision tasks reformulated into multiple choice questions with visual support.
What is the size of the dataset?
The dataset contains 3,807 examples for approximately 797 MB.
Is this dataset suitable for training and evaluating multimodal models?
Yes, BLINK is designed specifically to test and train multimodal LLMs on tasks requiring fine visual perception.
