Jedi: Vision-Language Data with Absolute Coordinates for Qwen2.5-VL

Created by xlangai and updated in June 2025, this dataset provides image-to-text data specifically formatted for the Qwen2.5-VL vision-language model. It utilizes absolute coordinates for spatial grounding tasks and is associated with research papers Arxiv:2505.13227 and Arxiv:2502.13923.

Use Cases

• Training Qwen2.5-VL models for spatial grounding using the absolute coordinate system
• Fine-tuning vision-language models for image-to-text tasks requiring precise pixel-to-token mapping
• Evaluating model performance on tasks involving resized image inputs within the 2700x14x14x2x2 pixel constraint

Strengths

• Optimized for Qwen2.5-VL architecture
• Includes absolute coordinate labels for spatial reasoning
• Released under Apache 2.0 license

Limitations

• Requires strict adherence to specific image resizing logic to maintain coordinate alignment
• Lack of public documentation on total record count

Provenance

Source

xlangai

Freshness

Last updated June 27, 2025.