SpatialBench evaluates 41 spatial foundation models across 6 paradigms and 5 task suites, finds they are not all-round players, and introduces the DA-Next-5M dataset plus DA-Next baseline model.
hub Canonical reference
TTT3R: 3D Reconstruction as Test-Time Training
Canonical reference. 100% of citing Pith papers cite this work as background.
35
Pith papers citing it
Background
100% of classified citations
abstract
Modern Recurrent Neural Networks have become a competitive architecture for 3D reconstruction due to their linear-time complexity. However, their performance degrades significantly when applied beyond the training context length, revealing limited length generalization. In this work, we revisit the 3D reconstruction foundation models from a Test-Time Training perspective, framing their designs as an online learning problem. Building on this perspective, we leverage the alignment confidence between the memory state and incoming observations to derive a closed-form learning rate for memory updates, to balance between retaining historical information and adapting to new observations. This training-free intervention, termed TTT3R, substantially improves length generalization, achieving a $2\times$ improvement in global pose estimation over baselines, while operating at 20 FPS with just 6 GB of GPU memory to process thousands of images. Code is available in https://rover-xingyu.github.io/TTT3R
hub tools
citation-role summary
background 6
citation-polarity summary
roles
background 6polarities
background 6representative citing papers
DrivingDepth achieves SOTA metric depth on nuScenes by residual pixel-wise scale correction on frozen foundation models using sparse LiDAR prompts, preserving geometric consistency.
Mamba-VGGT introduces a Sliding Window Mamba memory module and Zero-Init Spatial Memory Injector to enable persistent long-range geometric reasoning in VGGT for extended video sequences.
A feature-free monocular VINS initialization method that uses feed-forward 3D model point cloud predictions achieves over 90% success rate with under 1.2 seconds of data and performs robustly in degraded environments.
GlobalSplat achieves competitive novel-view synthesis on RealEstate10K and ACID using only 16K Gaussians via global scene tokens and coarse-to-fine training, with a 4MB footprint and under 78ms inference.
TokenGS uses learnable Gaussian tokens in an encoder-decoder architecture to regress 3D means directly, achieving SOTA feed-forward reconstruction on static and dynamic scenes with better robustness.
RobotPan predicts metric-scaled compact 3D Gaussians from calibrated multi-view inputs via spherical coordinates and hierarchical voxel priors for real-time 360° robotic perception and reconstruction.
AnyImageNav uses a semantic-to-geometric cascade with 3D multi-view foundation models to recover precise 6-DoF poses from goal images, achieving 0.27m position error and state-of-the-art success rates on Gibson and HM3D benchmarks.
Test-time constrained optimization incorporates priors into pre-trained multiview transformers via self-supervised losses and penalty terms to improve 3D reconstruction accuracy.
ZipMap achieves linear-time bidirectional 3D reconstruction by zipping image collections into a compact stateful representation via test-time training layers.
MoonSeg3R is the first method for online monocular 3D instance segmentation, achieving performance competitive with RGB-D systems by using CUT3R priors for geometric consistency and temporal query memory.
A real multi-city, multi-kilometer surround-view driving dataset plus an urban-tailored 3DGS baseline shows that city-scale reconstruction still degrades with scale, off-trajectory views, and real-world noise.
Argus plus Realsee3D deliver state-of-the-art metric camera pose, depth, and point-cloud reconstruction from unordered indoor panoramas via learned covisibility anchoring and geometric factorization.
Anchor3R reframes feed-forward 3D reconstruction as current-centric local measurement prediction, using loop-closure and motion averaging to produce coherent global maps from visual streams.
Robust Dreamer uses Latent Gaussian Memory anchored to diffusion latents and Deviation Learning with a Dynamic Deviation Archive to reduce drift in long-horizon action-controlled image-to-video generation, reporting SOTA results on ScanNet, DL3DV, and OmniWorldGame.
FOUND-IT constructs evolving task-driven 3D scene graphs with on-demand granularity from monocular cameras by augmenting foundation models, reporting 79% higher accuracy on a grounding benchmark and real-time Jetson deployment.
DyFN is a lightweight recurrent module that dynamically normalizes latent feature statistics to remove scale-shift drift and achieve state-of-the-art temporal consistency in streaming monocular geometry estimation while updating only 2% of parameters.
UniT unifies online and offline 3D geometry perception via a Group Autoregressive Transformer that processes observation groups with anchor-free point map prediction and a scale-adaptive loss.
LongDPM introduces an overlap-aware chunk-based framework that registers and fuses local dynamic reconstructions to achieve coherent long-range 4D geometry and tracking from monocular video.
A closed-form scalar frame-level gate α_t derived from internal feature changes extends effective memory in recurrent 3D reconstruction and improves accuracy on long sequences up to 4541 frames.
RetrieveVGGT enables constant-memory long-context streaming 3D reconstruction by retrieving relevant frames via query-key similarities in VGGT's first attention layer, outperforming StreamVGGT and others.
Spark3R achieves up to 28x speedup on 1000-frame 3D reconstruction inputs by asymmetrically reducing query and key-value tokens in Vision Transformers while keeping competitive quality.
The paper proposes ray-aware pointer memory with adaptive retain-or-replace updates to improve long-term stability and pose accuracy in streaming 3D reconstruction.
Converts pretrained Vision Transformers to linear-complexity TTT models via architectural and representational alignment, demonstrated by linearizing Stable Diffusion 3.5 with 1-hour fine-tuning to match quality at 1.32-1.47x faster inference.
citing papers explorer
-
SpatialBench: Is Your Spatial Foundation Model an All-Round Player?
SpatialBench evaluates 41 spatial foundation models across 6 paradigms and 5 task suites, finds they are not all-round players, and introduces the DA-Next-5M dataset plus DA-Next baseline model.
-
DrivingDepth: Sparse-Prompted Pixel-wise Scale Correction for Driving Depth Estimation
DrivingDepth achieves SOTA metric depth on nuScenes by residual pixel-wise scale correction on frozen foundation models using sparse LiDAR prompts, preserving geometric consistency.
-
Mamba-VGGT: Persistent Long-Sequence Video Geometry Grounded Transformer via External Sliding Window Mamba Memory
Mamba-VGGT introduces a Sliding Window Mamba memory module and Zero-Init Spatial Memory Injector to enable persistent long-range geometric reasoning in VGGT for extended video sequences.
-
Efficient Feature-Free Initialization for Monocular Visual-Inertial Systems Using a Feed-Forward 3D Model
A feature-free monocular VINS initialization method that uses feed-forward 3D model point cloud predictions achieves over 90% success rate with under 1.2 seconds of data and performs robustly in degraded environments.
-
GlobalSplat: Efficient Feed-Forward 3D Gaussian Splatting via Global Scene Tokens
GlobalSplat achieves competitive novel-view synthesis on RealEstate10K and ACID using only 16K Gaussians via global scene tokens and coarse-to-fine training, with a 4MB footprint and under 78ms inference.
-
TokenGS: Decoupling 3D Gaussian Prediction from Pixels with Learnable Tokens
TokenGS uses learnable Gaussian tokens in an encoder-decoder architecture to regress 3D means directly, achieving SOTA feed-forward reconstruction on static and dynamic scenes with better robustness.
-
RobotPan: A 360$^\circ$ Surround-View Robotic Vision System for Embodied Perception
RobotPan predicts metric-scaled compact 3D Gaussians from calibrated multi-view inputs via spherical coordinates and hierarchical voxel priors for real-time 360° robotic perception and reconstruction.
-
AnyImageNav: Any-View Geometry for Precise Last-Meter Image-Goal Navigation
AnyImageNav uses a semantic-to-geometric cascade with 3D multi-view foundation models to recover precise 6-DoF poses from goal images, achieving 0.27m position error and state-of-the-art success rates on Gibson and HM3D benchmarks.
-
Learning 3D Reconstruction with Priors in Test Time
Test-time constrained optimization incorporates priors into pre-trained multiview transformers via self-supervised losses and penalty terms to improve 3D reconstruction accuracy.
-
ZipMap: Linear-Time Stateful 3D Reconstruction via Test-Time Training
ZipMap achieves linear-time bidirectional 3D reconstruction by zipping image collections into a compact stateful representation via test-time training layers.
-
MoonSeg3R: Monocular Online Zero-Shot Segment Anything in 3D with Reconstructive Foundation Priors
MoonSeg3R is the first method for online monocular 3D instance segmentation, achieving performance competitive with RGB-D systems by using CUT3R priors for geometric consistency and temporal query memory.
-
WildCity: A Real-World City-Scale Testbed for Rendering, Simulation, and Spatial Intelligence
A real multi-city, multi-kilometer surround-view driving dataset plus an urban-tailored 3DGS baseline shows that city-scale reconstruction still degrades with scale, off-trajectory views, and real-world noise.
-
Argus: Metric Panoramic 3D Reconstruction for Indoor Scenes
Argus plus Realsee3D deliver state-of-the-art metric camera pose, depth, and point-cloud reconstruction from unordered indoor panoramas via learned covisibility anchoring and geometric factorization.
-
Anchor3R: Streaming 3D Reconstruction with Transient Anchors for Long-Horizon Visual Mapping
Anchor3R reframes feed-forward 3D reconstruction as current-centric local measurement prediction, using loop-closure and motion averaging to produce coherent global maps from visual streams.
-
Robust Dreamer: Deviation-Aware Latent Gaussian Memory for Action-Controlled AR Video Generation
Robust Dreamer uses Latent Gaussian Memory anchored to diffusion latents and Deviation Learning with a Dynamic Deviation Archive to reduce drift in long-horizon action-controlled image-to-video generation, reporting SOTA results on ScanNet, DL3DV, and OmniWorldGame.
-
FOUND-IT: Foundation-model-first Task-driven 3D Scene Graphs with Granularity on Demand
FOUND-IT constructs evolving task-driven 3D scene graphs with on-demand granularity from monocular cameras by augmenting foundation models, reporting 79% higher accuracy on a grounding benchmark and real-time Jetson deployment.
-
Stabilizing Streaming Video Geometry via Dynamic Feature Normalization
DyFN is a lightweight recurrent module that dynamically normalizes latent feature statistics to remove scale-shift drift and achieve state-of-the-art temporal consistency in streaming monocular geometry estimation while updating only 2% of parameters.
-
UniT: Unified Geometry Learning with Group Autoregressive Transformer
UniT unifies online and offline 3D geometry perception via a Group Autoregressive Transformer that processes observation groups with anchor-free point map prediction and a scale-adaptive loss.
-
LongDPM: Overlap-Aware 4D Reconstruction from Long Monocular Videos
LongDPM introduces an overlap-aware chunk-based framework that registers and fuses local dynamic reconstructions to achieve coherent long-range 4D geometry and tracking from monocular video.
-
Rethinking the State Update Gate for Long-Sequence Recurrent 3D Reconstruction
A closed-form scalar frame-level gate α_t derived from internal feature changes extends effective memory in recurrent 3D reconstruction and improves accuracy on long sequences up to 4541 frames.
-
Attention Itself Could Retrieve.RetrieveVGGT: Training-Free Long Context Streaming 3D Reconstruction via Query-Key Similarity Retrieval
RetrieveVGGT enables constant-memory long-context streaming 3D reconstruction by retrieving relevant frames via query-key similarities in VGGT's first attention layer, outperforming StreamVGGT and others.
-
Spark3R: Asymmetric Token Reduction Makes Fast Feed-Forward 3D Reconstruction
Spark3R achieves up to 28x speedup on 1000-frame 3D reconstruction inputs by asymmetrically reducing query and key-value tokens in Vision Transformers while keeping competitive quality.
-
Ray-Aware Pointer Memory with Adaptive Updates for Streaming 3D Reconstruction
The paper proposes ray-aware pointer memory with adaptive retain-or-replace updates to improve long-term stability and pose accuracy in streaming 3D reconstruction.
-
Linearizing Vision Transformer with Test-Time Training
Converts pretrained Vision Transformers to linear-complexity TTT models via architectural and representational alignment, demonstrated by linearizing Stable Diffusion 3.5 with 1-hour fine-tuning to match quality at 1.32-1.47x faster inference.
-
Feed-Forward 3D Scene Modeling: A Problem-Driven Perspective
The paper proposes a problem-driven taxonomy for feed-forward 3D scene modeling that groups methods by five core challenges: feature enhancement, geometry awareness, model efficiency, augmentation strategies, and temporal-aware modeling.
-
Scal3R: Scalable Test-Time Training for Large-Scale 3D Reconstruction
Scal3R achieves better accuracy and consistency in large-scale 3D scene reconstruction by maintaining a compressed global context through test-time adaptation of lightweight neural networks on long video sequences.
-
OVGGT: O(1) Constant-Cost Streaming Visual Geometry Transformer
OVGGT achieves constant O(1) memory and compute for streaming 3D geometry reconstruction by using FFN-residual-based KV cache compression and dynamic anchor protection, matching state-of-the-art accuracy on long sequences.
-
ViT$^3$: Unlocking Test-Time Training in Vision
ViT³ is a Test-Time Training vision model that achieves linear complexity, matches or exceeds other linear models like Mamba on classification, generation, detection and segmentation, and narrows the gap to standard vision Transformers.
-
LIST3R: Long-sequence Instance-aware 3D Reconstruction
LIST3R reconnects fragmented video subsequences using persistent instance anchors with semantic and geometric evidence to produce consistent global 3D reconstructions.
-
$R^3$: 3D Reconstruction via Relative Regression
R³ uses relative regression with confidence-weighted constraints from an MLP to support long-context offline and streaming 3D reconstruction without global coordinate assumptions.
-
HorizonStream: Long-Horizon Attention for Streaming 3D Reconstruction
HorizonStream is a long-horizon Transformer that factorizes geometric evidence influence into channel-wise linear attention for long-range temporal propagation and local spatiotemporal attention for short-range matching, claiming stable generalization from 48-frame training to over 10,000-frame test
-
IVGT: Implicit Visual Geometry Transformer for Neural Scene Representation
IVGT implicitly models continuous neural scene representations from pose-free multi-view images to enable coherent surface extraction, novel view synthesis, and related 3D tasks via SDF and color prediction.
-
StreamCacheVGGT: Streaming Visual Geometry Transformers with Robust Scoring and Hybrid Cache Compression
StreamCacheVGGT improves streaming 3D geometry reconstruction accuracy and stability under fixed memory by using cross-layer token importance scoring and hybrid cache compression instead of pure eviction.
-
FrameVGGT: Coherence-Preserving Memory for Bounded Streaming Geometry
FrameVGGT maintains stable long-horizon 3D reconstruction, depth, and pose under fixed memory by organizing history as complementary frame-wise KV prototypes plus sparse anchors.
- Stream3D: Sequential Multi-View 3D Generation via Evidential Memory