Explained: LingBot-Vision's Masked Boundary Modeling
Illustrious-Data1712 · reddit · 2026-07-07
A deeper look into Robbyant's open-sourced LingBot-Vision: while its architecture belongs to the DINO lineage, it introduces "Masked Boundary Modeling." In this approach, the teacher predicts dense boundary fields online, and tokens carrying boundaries are forcibly included in the student mask. The boundary field is converted into a pixel-wise class distribution to stabilize self-distillation, and segmentation results are verified via "proof by contradiction"—entirely without labels, text supervision, or external edge detectors.
Training used only 161 million curated images, less than a third of DINOv3's training samples. Internal tests show the 1.1B ViT-g achieves an RMSE of 0.296 on NYUv2 depth estimation, beating DINOv3-7B, while the distilled ViT-L matches DINOv3-7B with about 1/23 the parameters. Weights are directly available on HuggingFace and GitHub.
Related event: Ant Robbyant Open-Sources LingBot Vision Models, Topping Depth Benchmarks(18 posts)→
More from Models
- Kimi K3.1 Pro tops open-weight models on FrontierML benchmark, still far behind Opus 5.5 — teortaxesTex · 2026-10-27
- Open Models Match GPT-6 on MedMCQA but Trail 17 Points on Hard Clinical Cases — keunwoochoi · 2026-10-27
- Health-Focused Benchmark Reranks LLMs: Muse Spark Open Jumps from #9 to #1 — keunwoochoi · 2026-10-27
- DeepSeek cut KV cache per token from 389KB to 890B in two years as open-closed gap debate rages — teortaxesTex · 2026-10-27
- GPT-6 Astra controls dexterous hands zero-shot at just 5.2% success in 500 trials — stingning · 2026-10-27
- Unverified: fable 5.5 speculated at 4-5T parameters to outscale 1-3T Opus — haider1 · 2026-10-27