LibreLingBotVision family
-------------------------

The LingBot-Vision ViT backbone in this directory is a native port of
lingbot-vision (https://github.com/robbyant/lingbot-vision, commit
151e463), released by
Robbyant (Ant Group) under the Apache License, Version 2.0.
Copyright (c) 2026 Robbyant.
Reference: "Vision Pretraining for Dense Spatial Perception",
Fu et al., 2026, arXiv:2607.05247.

Pretrained backbone weights are distributed by Robbyant under Apache-2.0
(https://huggingface.co/collections/robbyant/lingbot-vision). LibreYOLO-hosted
LibreLingBotVision*-sem.pt checkpoints combine those backbone weights with a
dense head trained by LibreYOLO; see each weight repo's README for the
training data.

Architectural-lineage note: the upstream release documents its ViT as building
on DINOv2/DINOv3 (Meta AI). The upstream repository distributes that code
under Apache-2.0, which is the license this port inherits; the port was made
from the Robbyant repository only. The architecture family (RoPE, register
tokens, masked-K bias) is the one published in the DINOv3 report.
