Good Token Hunting: A Hitchhiker's Guide to Token Selection for Visual Geometry Transformers
cs.CV, cs.AI, cs.GR, cs.LG, cs.RO
Submitted: 2026-05-22
Updated: 2026-11-07
Code: https://github.com/facebookresearch/vggt
Project page: https://zsh2000.github.io/good-token-hunting.github.io
Terminology
Sources
- DePT3R: Joint Dense Point Tracking and 3D Reconstruction of Dynamic Scenes in a Single Forward Pass
- Geometric Context Transformer for Streaming 3D Reconstruction
- StereoVGGT: A Training-Free Visual Geometry Transformer for Stereo Vision
- Keyframe-Based Feed-Forward Visual Odometry
- Reloc-VGGT: Visual Re-localization with Geometry Grounded Transformer
- Building temporally coherent 3D maps with VGGT for memory-efficient Semantic SLAM
- MeMix: Writing Less, Remembering More for Streaming 3D Reconstruction
- SceneVGGT: VGGT-based online 3D semantic SLAM for indoor scene understanding and navigation
- DynamicVGGT: Learning Dynamic Point Maps for 4D Scene Reconstruction in Autonomous Driving
- VGGT4D: Mining Motion Cues in Visual Geometry Transformers for 4D Scene Reconstruction
- DriveVGGT: Calibration-Constrained Visual Geometry Transformers for Multi-Camera Autonomous Driving
- FILT3R: Latent State Adaptive Kalman Filter for Streaming 3D Reconstruction
- G-CUT3R: Guided 3D Reconstruction with Camera and Depth Prior Integration
- Distill3R: A Pipeline for Democratizing 3D Foundation Models on Commodity Hardware
- Analyzing the Mechanism of Attention Collapse in VGGT from a Dynamics Perspective
- Mem3R: Streaming 3D Reconstruction with Hybrid Memory
- StreamCacheVGGT: Streaming Visual Geometry Transformers with Robust Scoring and Hybrid Cache Compression
- VGGT-X: When VGGT Meets Dense Novel View Synthesis
- OVGGT: O(1) Constant-Cost Streaming Visual Geometry Transformer
- UniScale: Unified Scale-Aware 3D Reconstruction for Multi-View Understanding via Prior Injection for Robotic Perception
Related papers
- TEA: Text Encoder Alignment for Robust Concept Erasure in Text-to-Image Models
- Do Egocentric Video-Language Models Capture Both Hand- and Object-Centric Cues?
- Aura: Consistent Multi-Subject Video Generation via VLM-Grounded Semantic Alignment
- PVRF: All-in-one Adverse Weather Removal via Prior-modulated and Velocity-constrained Rectified Flow
- DLWM: Diverse Latent World Models for Efficient Multimodal Reasoning
- Out of Sight, Still in Mind: Token Compression for Omni-LLMs