الملخص
تمت ترجمة أجزاء من هذه الصفحة آلياً وقد تحتوي على أخطاء.
Spatial reasoning with metric constraints requires linking objects to geometric measurements and preserving their numerical content during language reasoning. We present DepthEvidence, a 4B model that uses its own dense metric predictions as object-grounded evidence for language generation. A camera-conditioned decoder predicts full-resolution metric depth using multi-scale visual features and high-resolution RGB refinement. A dense-to-language interface converts predicted depths and decoder features into object-aligned continuous geometry tokens anchored to object identifiers. Geometric supervision encourages metric information to remain recoverable before and after language-context interaction, while instruction tuning supports object measurement and compositional reasoning. We introduce a Depth-VQA benchmark evaluating object-depth queries, relative comparisons, and decisions combining spatial and numerical constraints. Across nine datasets, DepthEvidence achieves the highest average dense $δ_1$ among evaluated methods, competitive with specialized estimators. It also leads the evaluated methods in instance-level metric depth estimation and overall accuracy on both relative and metric reasoning tracks, while broadly preserving general VQA performance and improving spatial understanding relative to the base model.
الكلمات المفتاحية
الموضوع
بيانات النشر
- المجلة
- غير متاح
- وصول مفتوح
- وصول مفتوح أخضر
اقتبس هذه المقالة
APA 7
Wei, J., Yao, Y., Cui, M., Li, M., Zhong, H., Bai, S., & Yang, Z. (2026). DepthEvidence: توحيد التنبؤ بالعمق المتري والاستدلال الهندسي في نماذج اللغة متعددة الوسائط. https://omanscience.com/ar/articles/depthevidence-unifying-metric-depth-prediction-and-geometric-reasoning-in-multimodal-language-models
MLA 9
Wei, Jiangning, et al. "DepthEvidence: توحيد التنبؤ بالعمق المتري والاستدلال الهندسي في نماذج اللغة متعددة الوسائط." https://omanscience.com/ar/articles/depthevidence-unifying-metric-depth-prediction-and-geometric-reasoning-in-multimodal-language-models.
شيكاغو (المؤلف–التاريخ)
Wei, Jiangning, Yuan Yao, Miaomiao Cui, Mingsheng Li, Humen Zhong, Shuai Bai, and Zhibo Yang. 2026. "DepthEvidence: توحيد التنبؤ بالعمق المتري والاستدلال الهندسي في نماذج اللغة متعددة الوسائط." https://omanscience.com/ar/articles/depthevidence-unifying-metric-depth-prediction-and-geometric-reasoning-in-multimodal-language-models.
هارفارد
Wei, J., Yao, Y., Cui, M., Li, M., Zhong, H., Bai, S. and Yang, Z. (2026) 'DepthEvidence: توحيد التنبؤ بالعمق المتري والاستدلال الهندسي في نماذج اللغة متعددة الوسائط', Available at: https://omanscience.com/ar/articles/depthevidence-unifying-metric-depth-prediction-and-geometric-reasoning-in-multimodal-language-models.
فانكوفر
Wei J, Yao Y, Cui M, Li M, Zhong H, Bai S, et al. DepthEvidence: توحيد التنبؤ بالعمق المتري والاستدلال الهندسي في نماذج اللغة متعددة الوسائط. https://omanscience.com/ar/articles/depthevidence-unifying-metric-depth-prediction-and-geometric-reasoning-in-multimodal-language-models
IEEE
J. Wei, Y. Yao, M. Cui, M. Li, H. Zhong, S. Bai, and Z. Yang, "DepthEvidence: توحيد التنبؤ بالعمق المتري والاستدلال الهندسي في نماذج اللغة متعددة الوسائط," https://omanscience.com/ar/articles/depthevidence-unifying-metric-depth-prediction-and-geometric-reasoning-in-multimodal-language-models.