mlsanityData to 7 Oct 2026

FrameVGGT: Coherence-Preserving Memory for Bounded Streaming Geometry

arXiv 2603.07690 ↗First posted 8 Mar 2026cs.CVcited by 2

On the map: Efficient LLM inference and serving → Token pruning and merging for ViTs

Abstract (arXiv)

Streaming Visual Geometry Transformers such as StreamVGGT enable strong online 3D perception, but their KV-cache grows unbounded over long streams, limiting practical deployment. We study bounded-memory streaming geometry from the perspective of memory organization: unlike language modeling, where useful information can often be compressed at token level, geometry-driven inference relies on coherent and mutually compatible observations across views. Under fixed memory budgets, retaining history as isolated entries can progressively fragment the geometric context needed for stable long-horizon matching and fusion. We therefore propose FrameVGGT, a bounded-memory framework that maintains a fixed-capacity set of complementary memory units for streaming geometry. In our implementation, each unit is instantiated as a frame-wise KV segment summarized by a compact key-space prototype, together with a sparse anchor tier for persistent long-range references. Across long-sequence 3D reconstruction, video depth estimation, and camera pose estimation, FrameVGGT achieves favorable accuracy--memory trade-offs under bounded budgets while maintaining more stable geometry over long streams.

Key facts. Cited by 2 papers in the map (data through 7 Oct 2026). Placed in Token pruning and merging for ViTs, one of 12 directions in Efficient LLM inference and serving. Builds on 28 of its 42 references that are in the map. First posted on arXiv 8 Mar 2026.

Builds on

28 of 42 references.

Closest ancestors · 2

Same direction, newest first

  1. 2026XStreamVGGT: Extremely Memory-Efficient Streaming Vision Geometry Grounded Transformer with KV Cache Compression3 Jan
  2. 2025Evict3R: Training-Free Token Eviction for Memory-Bounded Streaming Visual Geometry Transformers22 Sep

Other references · 26

  1. 2023DUSt3R: Geometric 3D Vision Made Easycited by 1,526
  2. 2025VGGT: Visual Geometry Grounded Transformercited by 1,475
  3. 2015ORB-SLAM: a Versatile and Accurate Monocular SLAM Systemcited by 983
  4. 2024SnapKV: LLM Knows What You are Looking for Before Generationcited by 848
  5. 2023H2O: Heavy-Hitter Oracle for Efficient Generative Inference of Large Language Modelscited by 805
  6. 2025Depth Anything 3: Recovering the Visual Space from Any Viewscited by 626
  7. 2025Continuous 3D Perception Model with Persistent Statecited by 527
  8. 2023Scissorhands: Exploiting the Persistence of Importance Hypothesis for LLM KV Cache Compression at Test Timecited by 428
  9. 2021Neural RGB-D Surface Reconstructioncited by 312
  10. 2025MapAnything: Universal Feed-Forward Metric 3D Reconstructioncited by 295
  11. 2025π3: Permutation-Equivariant Visual Geometry Learningcited by 292
  12. 20243D Reconstruction with Spatial Memorycited by 269
14 more
  1. 2019ReFusion: 3D Reconstruction in Dynamic Environments for RGB-D Cameras Exploiting Residualscited by 180
  2. 2024MASt3R-SfM: a Fully-Integrated Solution for Unconstrained Structure-from-Motioncited by 161
  3. 2025Streaming 4D Visual Geometry Transformercited by 147
  4. 2025VGGT-SLAM: Dense RGB SLAM Optimized on the SL(4) Manifoldcited by 130
  5. 2025TTT3R: 3D Reconstruction as Test-Time Trainingcited by 122
  6. 2025VGGT-Long: Chunk it, Loop it, Align it -- Pushing VGGT's Limits on Kilometer-scale Long RGB Sequencescited by 107
  7. 2024SLAM3R: Real-Time Dense Scene Reconstruction from Monocular RGB Videoscited by 100
  8. 2025Point3R: Streaming 3D Reconstruction with Explicit Spatial Pointer Memorycited by 82
  9. 2026InfiniteVGGT: Visual Geometry Grounded Transformer for Endless Streamscited by 51
  10. 2025LONG3R: Long Sequence Streaming 3D Reconstructioncited by 45
  11. 2025WinT3R: Window-Based Streaming Reconstruction with Camera Token Poolcited by 32
  12. 2019Robust Dense Mapping for Large-Scale Dynamic Environmentscited by 21
  13. 2025SAIL-Recon: Large SfM by Augmenting Scene Regression with Localizationcited by 18
  14. 2017Sparse-then-Dense Alignment based 3D Map Reconstruction Method for Endoscopic Capsule Robotscited by 3

Built on by

Cited by year (2026 to 7 Oct)

Citing papers per year: 2026: 2.

2 papers.

  1. 2026FFVO: A Feedforward Pose Decoder for Long-Horizon Visual Odometrycited by 0
  2. 2026Slot3R: Set-Associative Spatial Memory for Streaming 3D Reconstructioncited by 0

Similar, not cited

Close in wording, no citation either way

  1. 2026OVGGT: O(1) Constant-Cost Streaming Visual Geometry Transformer
  2. 2026XStreamVGGT: Extremely Memory-Efficient Streaming Vision Geometry Grounded Transformer with KV Cache Compression
  3. 2026Geometry-Aware Implicit Memory for Video World Models
  4. 2026Attention Itself Could Retrieve.RetrieveVGGT: Training-Free Long Context Streaming 3D Reconstruction via Query-Key Similarity Retrieval
  5. 2025LiteVGGT: Boosting Vanilla VGGT via Geometry-aware Cached Token Merging
  6. 2026Stream3D: Sequential Multi-View 3D Generation via Evidential Memory
  7. 2026LongStream: Long-Sequence Streaming Autoregressive Visual Geometry
  8. 2026GHOST: Geometry-Hierarchical Online Streaming Token Eviction for Efficient 3D Reconstruction