Mini toolsLingBot Vision Giant
By Robbyant
LingBot-Vision-Giant is the largest backbone in the LingBot-Vision family, built on a ViT-g/16 architecture and trained from random initialization using self-supervised teacher-student pretraining. During training, teacher-discovered boundary tokens are forced into the masked set and receive both semantic self-distillation and categorical boundary-field supervision, yielding dense patch representations that preserve boundaries, shapes, and semantic regions. This repository provides a backbone-only PyTorch checkpoint intended for inference, feature extraction, PCA visualization, and downstream dense prediction research. Direct uses include dense feature visualization and image feature extraction; downstream uses include depth estimation, semantic segmentation, video object segmentation, and backbone initialization for dense prediction models. Only backbone weights are included; training-time heads and optimizer states are excluded.
New
Image
Overview
LingBot-Vision-Giant is a ViT-g/16 self-supervised Vision Transformer backbone for dense visual perception. Pretrained with masked boundary modeling, a boundary-centric objective that produces spatially structured patch features while retaining strong semantic representations, intended for feature extraction and dense prediction research.
Last updated: July 8, 2026
