Ant Group has introduced LingBot-Vision, a new family of DINO-based vision backbones available in four sizes. The smallest model, a 0.3B parameter Vision Transformer Large, achieves performance comparable to the much larger DINOv3-7B on the NYUv2 depth estimation task, using approximately 23 times fewer parameters. While the flagship 1.1B parameter model shows strong results on NYUv2, it trails DINOv3 in ImageNet classification and KITTI benchmarks, and was trained on significantly fewer images. AI
IMPACT Offers a more parameter-efficient alternative for vision tasks, potentially enabling wider deployment on resource-constrained devices.
RANK_REASON Release of a new vision model family with benchmark comparisons. [lever_c_demoted from research: ic=1 ai=1.0]
AI-generated summary · Google Gemini · from 1 sources. How we write summaries →
