Pith. sign in

REVIEW 2 major objections 6 minor 37 references

A frozen foundation model plus a 3.83M fast path can deliver real-time monocular depth by reusing spatial features across frames with only bounded accuracy loss.

Reviewed by Pith at T0; open to challenge. T0 means a machine referee read the full paper against a public rubric. the ladder, T0–T4 →

Challenge this review Re-run · record.json Download PDF Read on arXiv ↗

T0 review · grok-4.5

2026-07-14 23:35 UTC pith:QA6467C7

load-bearing objection Solid systems paper: amortize a frozen MDE foundation model with a 3.83M complementary-fusion fast path, real edge FPS, and measured lag-bounded degradation. the 2 major comments →

arxiv 2603.10438 v2 pith:QA6467C7 submitted 2026-03-11 cs.RO cs.CV

AsyncMDE: Real-Time Monocular Depth Estimation via Asynchronous Spatial Memory

Lianjie Ma , Yuquan Li , Bingzheng Jiang , Ziming Zhong , Han Ding , Lijun Zhu This is my paper
classification cs.RO cs.CV
keywords monocular depth estimationasynchronous inferencespatial memoryedge roboticsfoundation modelsreal-time perceptionfeature amortization
verification ladder T0 review T1 audit T2 compute T3 formal T4 reserved

The pith

A machine-rendered reading of the paper's core claim, the machinery that carries it, and where it could break.

AsyncMDE argues that monocular depth for continuous robot operation need not recompute a heavy foundation model on every frame. Adjacent viewpoints share most of their 3D structure, so the expensive model can run infrequently in the background, writing multi-scale features into a spatial memory, while a lightweight network runs at high rate in the foreground, fusing that memory with the current image and updating memory. With 3.83 million trainable parameters the fast path reaches 237 FPS on a desktop GPU and 161 FPS on a Jetson AGX Orin, recovering about 77 percent of the accuracy gap to the foundation model. Accuracy degrades smoothly and predictably between refreshes rather than collapsing, giving a practical route to foundation-quality depth on edge platforms without waiting for further model compression.

Core claim

Pairing a frozen depth foundation model as a slow path with a lightweight fast path linked by multi-scale spatial memory amortizes expensive scene representation over time. Complementary per-pixel fusion of cached foundation features with current observations, followed by autoregressive memory updates, produces depth estimates that stay close to the foundation model and degrade gracefully between refreshes, recovering most of the accuracy gap at roughly 25 times fewer trainable parameters.

What carries the argument

SpatialMemoryUnit: multi-scale cached features fused with the current lightweight encoder via a semantic gated modulation factor T (trust memory when static, inject new observation when changed), then written back autoregressively so foundation quality persists with controlled decay until the next slow-path refresh.

Load-bearing premise

Consecutive robot viewpoints change little enough that blending cached foundation features with the current frame in downsampled feature space, without pose or optical flow, keeps depth accurate until the next slow refresh.

What would settle it

On sustained large-motion sequences, check whether cycle-average accuracy for the intended refresh interval stays clearly above the standalone lightweight-encoder floor; if it collapses to that floor under realistic robot rates, the amortization claim fails for dynamic settings.

Watch this falsifier — get emailed when new claim-graph text bears on it.

If this is right

  • Edge robots can obtain near-foundation monocular depth at control rates without active sensors or full-backbone inference every frame.
  • System accuracy becomes a function of the hardware-determined fast/slow rate ratio and scales across platforms without retraining.
  • Task-specific external memory with complementary fusion beats pure distillation to a few million parameters and heavy general-purpose memory models for real-time depth.
  • The same amortization pattern extends to other dense perception tasks that rely on spatiotemporal continuity.
  • Fast–slow dual-process design can be applied at the perception layer itself, not only at decision-making.

Where Pith is reading between the lines

These are editorial extensions of the paper, not claims the author makes directly.

  • The same asynchronous memory pattern could amortize other heavy foundation outputs (normals, segmentation, flow) for continuous robot operation.
  • A motion-triggered slow-path refresh when the fraction of low-T pixels spikes would limit extreme-motion collapse without raising average compute.
  • Adding a lightweight temporal scale-alignment head would make the relative-depth system usable for metric navigation without redesigning the fusion core.
  • The encoder ablation (smaller encoder better) implies future fast paths should shrink further and act only as change detectors.

Editorial analysis

A structured set of objections, weighed in public.

Desk editor's note, referee report, simulated authors' rebuttal, and a circularity audit.

Referee Report

2 major / 6 minor

Summary. AsyncMDE proposes an asynchronous dual-path monocular depth system that amortizes a frozen Depth Anything V2-ViTB slow path over time via multi-scale spatial memory. A lightweight 3.83M-parameter fast path (MobileNetV3-Small encoder, SpatialMemoryUnit with semantic-gated complementary fusion, and inherited DPT RefineNet decoder) fuses cached foundation features with current observations, outputs depth, and autoregressively updates memory (Eqs. 9–10). Training uses DAv2 pseudo-labels and fixed refresh interval N=10 on NYUv2/TartanAir/BridgeData V2; evaluation reports cycle-average AbsRel/RMSE/δ1 on ScanNet, Bonn, and Sintel against real GT, lag-degradation curves (Fig. 3, N=20 OOD), encoder/SMU ablations (Tables IV–V), and Jetson AGX Orin TensorRT numbers (161 FPS). The central claim is that this recovers ~77% of the accuracy gap to DAv2-ViTB at 237 FPS (RTX 4090) with bounded, scene-dependent degradation between refreshes.

Significance. If the reported numbers hold, the work is a practically useful systems contribution for edge robot perception: it shows that foundation-model intermediate features can be amortized by a few-million-parameter complementary-fusion path without pose, flow, or warping, yielding real-time rates on Jetson Orin while remaining within a few points of DAv2-ViTB on indoor static/dynamic data. Strengths include multi-benchmark evaluation with real GT, explicit lag curves that match the convex-combination boundedness argument, ablations isolating memory initialization and gating, and concrete edge deployment measurements. The continuity assumption is stated and stress-tested (Sintel floor at FastPath-Only), so the result is falsifiable rather than overclaimed. The paradigm is transferable to other dense perception tasks that exploit spatiotemporal continuity.

major comments (2)
  1. Table II and §IV-B: the 77% gap-recovery claim and the primary accuracy–efficiency comparison rest on cycle averages over lag 0–9. Because accuracy is lag-dependent (Fig. 3), the paper should also report lag-0 (post-refresh) and lag-(N−1) (worst-in-cycle) metrics side-by-side with the cycle average for every method row, or at least for AsyncMDE vs. DAv2-ViTB and LiteMono†. Without that, readers cannot judge whether the amortized system is acceptable for control loops that care about worst-case depth within a refresh window.
  2. §III-C and Fig. 3 / Table V: the load-bearing continuity assumption (feature-space fusion without pose/flow/warping is sufficient) is acknowledged and bounded by the FastPath-Only floor, but the manuscript never quantifies how often large-scale T→0 occurs on real robot trajectories (e.g., fraction of pixels with T below a threshold, or fraction of frames where mean T drops below τ). Adding a short deployment-oriented statistic on BridgeData-style or real robot sequences would make the graceful-degradation claim more actionable for practitioners choosing N_eff.
minor comments (6)
  1. Abstract vs. body: abstract says “recovering 77% of the accuracy gap”; body Fig. 1 caption and §I use the same figure without stating the exact δ1 arithmetic (which baselines and which average). Spell out the formula once.
  2. Table I nomenclature lists T^(ℓ)_t ∈ (0,1) but the text also uses T′, T_final, T_L1, T_L4; a one-line clarification that T′ is the temporally smoothed version would reduce notation friction.
  3. §IV-B deployment: Neff ≈ 237/60 ≈ 4 on 4090 and ≈ 161/12 ≈ 13 on Orin TRT are useful; state whether the slow path is also TensorRT-optimized on Orin or only the fast path, since that affects the reported Neff.
  4. Fig. 4 qualitative comparison: least-squares alignment is noted; briefly state whether scale-shift is estimated per image or per sequence, for reproducibility of the visual comparison.
  5. Related work: FlashDepth and Buffer Anytime appear in the reference list but are only lightly positioned relative to the asynchronous amortization claim; a sentence on why they do not already solve the edge amortization problem would tighten the novelty framing.
  6. Typos / consistency: abstract “a foundation model and a lightweight model” vs. later “fast path”; ensure “AsyncMDE” vs. “AsyncMDE’s” possessives are uniform. Eq. (6) temperature k=4.0 is fixed without ablation—acceptable, but note it as a hyperparameter.

Circularity Check

0 steps flagged

No significant circularity: empirical systems result with external GT evaluation and independent baselines.

full rationale

AsyncMDE is an empirical systems paper, not a first-principles derivation. The load-bearing claims (3.83M fast path, 237/161 FPS, ~77% gap recovery, lag-bounded degradation) are measured against real depth GT on ScanNet, Bonn, and Sintel and compared to independent baselines (DAv2 variants, LiteMono†, VDA, CUT3R). Training on frozen DAv2 pseudo-labels and DAv2-initialized memory is standard teacher–student amortization, not a self-definitional loop: the reported AbsRel/RMSE/δ1 and lag curves (Fig. 3, Tables II–V) are external measurements, and FastPath-Only / encoder ablations establish an independent performance floor. The complementary-fusion form (Eqs. 9–10) is a design choice whose convex-combination boundedness is stated as a property of that design, then checked empirically—not a prediction forced by fitting the same quantity. No self-citation chain, uniqueness theorem from the authors, or renamed known result carries the central claim. Score 0 is appropriate.

Axiom & Free-Parameter Ledger

5 free parameters · 4 axioms · 3 invented entities

The central claim rests on a small set of engineering and domain premises plus a handful of hand-chosen scalars in the fusion and loss. No new physical entities are postulated; the invented pieces are architectural modules. The load-bearing domain assumption is spatiotemporal continuity of robot viewpoints; free parameters mainly control gating smoothness and training regularization rather than the headline accuracy numbers themselves.

free parameters (5)
  • refresh interval N (training) = 10 (train); 10/20 eval
    Fixed to 10 frames for training clips; evaluation uses N=10 cycle averages and N=20 lag curves. Directly sets the amortization schedule the accuracy claims depend on.
  • memory regularization threshold τ = 0.4
    Soft lower bound on mean T in L_mem; chosen as 0.4 to prevent early collapse to encoder-only behavior.
  • temporal smoothing β for T = 0.5
    EMA coefficient on modulation factors; set to 0.5 to suppress frame-to-frame T jitter.
  • semantic gate temperature k = 4.0
    Controls sharpness of switch between Layer-1 and Layer-4 T maps; set to 4.0.
  • loss weights (L_SSI, L_grad, L_mem) = 1.0, 0.5, 0.1
    Hand-set coefficients 1.0 / 0.5 / 0.1 that shape training trade-offs among accuracy, edges, and memory use.
axioms (4)
  • domain assumption Adjacent robot viewpoints share substantial 3D structure so temporal adaptation is far simpler than full scene representation from a single image.
    Stated in the introduction as the motivation for amortizing the foundation model; if false (e.g., pure random viewpoints), the fast path cannot reuse memory.
  • standard math Per-pixel convex combination O = T·M + (1−T)·F with T∈(0,1) and autoregressive Mt+1=Ot yields bounded fused features and exponentially decaying contribution of the initial memory.
    Section III-C; standard convex-combination stability argument used to claim graceful degradation for arbitrary N.
  • ad hoc to paper Fusion in 8×–32× downsampled feature space is robust enough to pixel-level motion that explicit pose estimation or optical-flow warping is unnecessary.
    Explicit design choice in III-C; trades warping failure modes for a predictable degradation rate when motion exceeds receptive-field tolerance.
  • domain assumption Frozen DAv2-ViTB multi-scale features form a high-quality spatial memory ceiling that a lightweight network can preserve and adapt.
    Initialization and slow-path design (Eq. 1, Fig. 2); accuracy claims are relative to this teacher.
invented entities (3)
  • SpatialMemoryUnit (complementary fusion + autoregressive multi-scale memory) no independent evidence
    purpose: Bridge slow foundation features and fast current observations so the lightweight path reuses rather than recomputes scene representation.
    Core architectural invention; no independent existence outside this system. Ablations show DAv2-initialized memory is essential versus encoder-only memory.
  • SemanticGatedModulator (dual-scale T with learnable semantic gate) no independent evidence
    purpose: Produce per-pixel trust maps that retain memory in static regions and inject observations in changed regions.
    Paper-specific gating design combining Layer-1 texture and Layer-4 semantic T; ablated against L1-only and L4-only variants.
  • AsyncMDE dual-path asynchronous perception system no independent evidence
    purpose: Amortize foundation-model cost over time with rate-controlled accuracy governed by hardware refresh ratio Neff.
    System-level construct pairing frozen slow path, lock-free feature cache, and fast path; the paper’s primary contribution framing.

pith-pipeline@v1.1.0-grok45 · 18463 in / 3707 out tokens · 40144 ms · 2026-07-14T23:35:14.086043+00:00 · methodology

0 comments
read the original abstract

Foundation-model-based monocular depth estimation offers a viable alternative to active sensors for robot perception, yet its computational cost often prohibits deployment on edge platforms. Existing methods perform independent per-frame inference, wasting the substantial computational redundancy between adjacent viewpoints in continuous robot operation. This paper presents AsyncMDE, an asynchronous depth perception system consisting of a frozen foundation model and a lightweight fast path that amortizes the foundation model's computational cost over time. The foundation model periodically produces high-quality spatial features in the background, while the lightweight fast path runs asynchronously in the foreground, fusing cached memory with current observations through complementary fusion, outputting depth estimates, and autoregressively updating memory. This enables cross-frame feature reuse with bounded accuracy degradation. With 3.83M trainable fast-path parameters and a 97.5M frozen slow path, AsyncMDE's fast path operates at 237 FPS on an RTX 4090, recovering 77% of the accuracy gap to the foundation model. Across indoor static, dynamic, and synthetic extreme-motion benchmarks, AsyncMDE degrades predictably and reaches 161 FPS fast-path inference on a TensorRT-optimized Jetson AGX Orin, supporting real-time edge deployment.

Figures

Figures reproduced from arXiv: 2603.10438 by Bingzheng Jiang, Han Ding, Lianjie Ma, Lijun Zhu, Yuquan Li, Ziming Zhong.

Figure 1
Figure 1. Figure 1: Overview of AsyncMDE. Top: the Slow Path (DAv2-ViTB) periodically refreshes spatial memory; the Fast Path fuses cached memory with each frame at high frequency, with depth maps at increasing lag showing graceful degradation. Bottom: efficiency–accuracy trade-off (three￾benchmark average δ1); AsyncMDE (3.83 M, 237 FPS) recovers 77% of the δ1 gap between the lightweight baseline and the foundation model. Bey… view at source ↗
Figure 2
Figure 2. Figure 2: AsyncMDE system overview. DAv2-ViTB runs asynchronously in the background (slow path, [PITH_FULL_IMAGE:figures/full_fig_p003_2.png] view at source ↗
Figure 3
Figure 3. Figure 3: Lag–accuracy degradation curves. The evaluation interval [PITH_FULL_IMAGE:figures/full_fig_p006_3.png] view at source ↗
Figure 4
Figure 4. Figure 4: Qualitative depth comparison (least-squares aligned). The three rows correspond to ScanNet (indoor static), Bonn (indoor dynamic), and Sintel [PITH_FULL_IMAGE:figures/full_fig_p007_4.png] view at source ↗
Figure 5
Figure 5. Figure 5: T-value visualization. Each row shows, from left to right, the refresh-frame RGB, T-value masks and depth estimates at lag=1 and lag=N. Top: indoor robotic manipulation; bottom: outdoor dynamic scene. Warm colors (T →1) indicate static regions; cool colors (T →0) indicate moving regions. Degradation primarily affects moving objects, while static structures maintain stable estimates even at high lag. E. Qua… view at source ↗

discussion (0)

Sign in with ORCID, Apple, or X to comment. Anyone can read and Pith papers without signing in.

Reference graph

Works this paper leans on

37 extracted references · 6 linked inside Pith

  1. [1]

    iPlanner: Imperative path planning,

    F. Yang, C. Wang, C. Cadena, and M. Hutter, “iPlanner: Imperative path planning,” inRobotics: Science and Systems XIX, 2023

    2023

  2. [2]

    ViPlanner: Visual semantic imperative learning for local navigation,

    P. Roth, J. Nubert, F. Yang, M. Mittal, and M. Hutter, “ViPlanner: Visual semantic imperative learning for local navigation,” in2024 IEEE International Conference on Robotics and Automation (ICRA), 2024, pp. 5243–5249

    2024

  3. [3]

    ViNT: A foundation model for visual navigation,

    D. Shah, A. Sridhar, N. Dashora, K. Stachowicz, K. Black, N. Hirose, and S. Levine, “ViNT: A foundation model for visual navigation,” in Proceedings of The 7th Conference on Robot Learning. PMLR, 2023, pp. 711–733, ISSN: 2640-3498

    2023

  4. [4]

    NoMaD: Goal masked diffusion policies for navigation and exploration,

    A. Sridhar, D. Shah, C. Glossop, and S. Levine, “NoMaD: Goal masked diffusion policies for navigation and exploration,” in2024 IEEE International Conference on Robotics and Automation (ICRA), 2024, pp. 63–70

    2024

  5. [5]

    A dual process VLA: Efficient robotic manipulation leveraging VLM,

    B. Han, J. Kim, and J. Jang, “A dual process VLA: Efficient robotic manipulation leveraging VLM,”arXiv preprint arXiv:2410.15549, 2024

    Pith/arXiv arXiv 2024

  6. [6]

    GR00T N1: An open foundation model for generalist humanoid robots,

    NVIDIA, J. Bjorck, F. Castañeda, N. Cherniadev, X. Da, R. Ding, L. J. Fan, Y . Fang, D. Fox, F. Hu, S. Huang, J. Jang, Z. Jiang, J. Kautz, K. Kundalia, L. Lao, Z. Li, Z. Lin, K. Lin, G. Liu, E. Llontop, L. Magne, A. Mandlekar, A. Narayan, S. Nasiriany, S. Reed, Y . L. Tan, G. Wang, Z. Wang, J. Wang, Q. Wang, J. Xiang, Y . Xie, Y . Xu, Z. Xu, S. Ye, Z. Yu,...

    Pith/arXiv arXiv 2025

  7. [7]

    OpenHelix: A short survey, empirical analysis, and open-source dual-system VLA model for robotic manipulation,

    C. Cui, P. Ding, W. Song, S. Bai, X. Tong, Z. Ge, R. Suo, W. Zhou, Y . Liu, B. Jia, H. Zhao, S. Huang, and D. Wang, “OpenHelix: A short survey, empirical analysis, and open-source dual-system VLA model for robotic manipulation,”arXiv preprint arXiv:2505.03912, 2025

    Pith/arXiv arXiv 2025

  8. [8]

    OpenVLA: An open-source vision-language-action model,

    M. J. Kim, K. Pertsch, S. Karamcheti, T. Xiao, A. Balakrishna, S. Nair, R. Rafailov, E. P. Foster, P. R. Sanketi, Q. Vuong, T. Kollar, B. Burchfiel, R. Tedrake, D. Sadigh, S. Levine, P. Liang, and C. Finn, “OpenVLA: An open-source vision-language-action model,” inProceedings of The 8th Conference on Robot Learning. PMLR, 2025, pp. 2679–2713, ISSN: 2640-3498

    2025

  9. [9]

    Towards depth foundation models: Recent trends in vision- based depth estimation,

    Z. Xu, H. Zhou, S. Peng, H. Lin, H. Guo, J. Shao, P. Yang, Q. Yang, S. Miao, X. He, Y . Wang, Y . Wang, R. Hu, Y . Liao, X. Zhou, and H. Bao, “Towards depth foundation models: Recent trends in vision- based depth estimation,”Computational Visual Media, pp. 1–29, 2026

    2026

  10. [10]

    Depth anything v2,

    L. Yang, B. Kang, Z. Huang, Z. Zhao, X. Xu, J. Feng, and H. Zhao, “Depth anything v2,” inAdvances in Neural Information Processing Systems, A. Globerson, L. Mackey, D. Belgrave, A. Fan, U. Paquet, J. Tomczak, and C. Zhang, Eds., vol. 37. Curran Associates, Inc., 2024, pp. 21 875–21 911

    2024

  11. [11]

    Metric3d v2: A versatile monocular geometric foundation model for zero-shot metric depth and surface normal estimation,

    M. Hu, W. Yin, C. Zhang, Z. Cai, X. Long, H. Chen, K. Wang, G. Yu, C. Shen, and S. Shen, “Metric3d v2: A versatile monocular geometric foundation model for zero-shot metric depth and surface normal estimation,”IEEE Transactions on Pattern Analysis and Machine Intelligence, vol. 46, no. 12, pp. 10 579–10 596, 2024

    2024

  12. [12]

    UniDepthV2: Universal monocular metric depth estimation made simpler,

    L. Piccinelli, C. Sakaridis, Y .-H. Yang, M. Segu, S. Li, W. Abbeloos, and L. Van Gool, “UniDepthV2: Universal monocular metric depth estimation made simpler,”IEEE Transactions on Pattern Analysis and Machine Intelligence, vol. 48, no. 3, pp. 2354–2367, 2026

    2026

  13. [13]

    Depth pro: Sharp monocular metric depth in less than a second,

    A. Bochkovskiy, A. Delaunoy, H. Germain, M. Santos, Y . Zhou, S. Richter, and V . Koltun, “Depth pro: Sharp monocular metric depth in less than a second,” inThe Thirteenth International Conference on Learning Representations, 2024

    2024

  14. [14]

    Learning temporally consistent video depth from video diffusion priors,

    J. Shao, Y . Yang, H. Zhou, Y . Zhang, Y . Shen, V . Guizilini, Y . Wang, M. Poggi, and Y . Liao, “Learning temporally consistent video depth from video diffusion priors,” inProceedings of the IEEE/CVF Confer- ence on Computer Vision and Pattern Recognition, 2025, pp. 22 841– 22 852

    2025

  15. [15]

    DepthCrafter: Generating consistent long depth sequences for open-world videos,

    W. Hu, X. Gao, X. Li, S. Zhao, X. Cun, Y . Zhang, L. Quan, and Y . Shan, “DepthCrafter: Generating consistent long depth sequences for open-world videos,” inProceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition, 2025, pp. 2005–2015

    2025

  16. [16]

    Video depth anything: Consistent depth estimation for super-long videos,

    S. Chen, H. Guo, S. Zhu, F. Zhang, Z. Huang, J. Feng, and B. Kang, “Video depth anything: Consistent depth estimation for super-long videos,” inProceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition, 2025, pp. 22 831–22 840

    2025

  17. [17]

    FlashDepth: Real-time streaming video depth estimation at 2k resolution,

    G. Chou, W. Xian, G. Yang, M. Abdelfattah, B. Hariharan, N. Snavely, N. Yu, and P. Debevec, “FlashDepth: Real-time streaming video depth estimation at 2k resolution,” inProceedings of the IEEE/CVF International Conference on Computer Vision, 2025, pp. 9638–9648

    2025

  18. [18]

    Buffer anytime: Zero-shot video depth and normal from image priors,

    Z. Kuang, T. Zhang, K. Zhang, H. Tan, S. Bi, Y . Hu, Z. Xu, M. Hasan, G. Wetzstein, and F. Luan, “Buffer anytime: Zero-shot video depth and normal from image priors,” inProceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition, 2025, pp. 17 660–17 670

    2025

  19. [19]

    Depth anything 3: Recovering the visual space from any views,

    H. Lin, S. Chen, J. Liew, D. Y . Chen, Z. Li, G. Shi, J. Feng, and B. Kang, “Depth anything 3: Recovering the visual space from any views,”arXiv preprint arXiv:2511.10647, 2025

    Pith/arXiv arXiv 2025

  20. [20]

    VGGT: Visual geometry grounded transformer,

    J. Wang, M. Chen, N. Karaev, A. Vedaldi, C. Rupprecht, and D. Novotny, “VGGT: Visual geometry grounded transformer,” in Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition, 2025, pp. 5294–5306

    2025

  21. [21]

    FastDepth: Fast monocular depth estimation on embedded systems,

    D. Wofk, F. Ma, T.-J. Yang, S. Karaman, and V . Sze, “FastDepth: Fast monocular depth estimation on embedded systems,” in2019 International Conference on Robotics and Automation (ICRA), 2019, pp. 6101–6108, ISSN: 2577-087X

    2019

  22. [22]

    Lite-mono: A lightweight CNN and transformer architecture for self-supervised monocular depth estimation,

    N. Zhang, F. Nex, G. V osselman, and N. Kerle, “Lite-mono: A lightweight CNN and transformer architecture for self-supervised monocular depth estimation,” inProceedings of the IEEE/CVF Confer- ence on Computer Vision and Pattern Recognition, 2023, pp. 18 537– 18 546

    2023

  23. [23]

    Kahneman,Thinking, Fast and Slow

    D. Kahneman,Thinking, Fast and Slow. Macmillan, 2011, google- Books-ID: SHvzzuCnuv8C

    2011

  24. [24]

    Fast-in-slow: A dual-system foundation model unifying fast manipulation within slow reasoning,

    H. Chen, J. Liu, C. Gu, Z. Liu, R. Zhang, X. Li, X. He, Y . Guo, C.-W. Fu, S. Zhang, and P.-A. Heng, “Fast-in-slow: A dual-system foundation model unifying fast manipulation within slow reasoning,” arXiv preprint arXiv:2506.01953, 2025

    Pith/arXiv arXiv 2025

  25. [25]

    Prompting depth anything for 4k resolution accurate metric depth estimation,

    H. Lin, S. Peng, J. Chen, S. Peng, J. Sun, M. Liu, H. Bao, J. Feng, X. Zhou, and B. Kang, “Prompting depth anything for 4k resolution accurate metric depth estimation,” inProceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition, 2025, pp. 17 070–17 080

    2025

  26. [26]

    Manipulation as in simulation: Enabling accurate geometry perception in robots,

    M. Liu, Z. Zhu, X. Han, P. Hu, H. Lin, X. Li, J. Chen, J. Xu, Y . Yang, Y . Lin, X. Li, Y . Yu, W. Zhang, T. Kong, and B. Kang, “Manipulation as in simulation: Enabling accurate geometry perception in robots,” arXiv preprint arXiv:2509.02530, 2025

    Pith/arXiv arXiv 2025

  27. [27]

    3d reconstruction with spatial memory,

    H. Wang and L. Agapito, “3d reconstruction with spatial memory,” in 2025 International Conference on 3D Vision (3DV), 2025, pp. 78–89, ISSN: 2475-7888

    2025

  28. [28]

    Continuous 3d perception model with persistent state,

    Q. Wang, Y . Zhang, A. Holynski, A. A. Efros, and A. Kanazawa, “Continuous 3d perception model with persistent state,” inProceed- ings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition, 2025, pp. 10 510–10 522

    2025

  29. [29]

    Searching for MobileNetV3,

    A. Howard, M. Sandler, G. Chu, L.-C. Chen, B. Chen, M. Tan, W. Wang, Y . Zhu, R. Pang, V . Vasudevan, Q. V . Le, and H. Adam, “Searching for MobileNetV3,” inProceedings of the IEEE/CVF International Conference on Computer Vision (ICCV), 2019

    2019

  30. [30]

    Indoor segmentation and support inference from RGBD images,

    N. Silberman, D. Hoiem, P. Kohli, and R. Fergus, “Indoor segmentation and support inference from RGBD images,” inComputer Vision – ECCV 2012, A. Fitzgibbon, S. Lazebnik, P. Perona, Y . Sato, and C. Schmid, Eds. Springer, 2012, pp. 746–760

    2012

  31. [31]

    TartanAir: A dataset to push the limits of visual SLAM,

    W. Wang, D. Zhu, X. Wang, Y . Hu, Y . Qiu, C. Wang, Y . Hu, A. Kapoor, and S. Scherer, “TartanAir: A dataset to push the limits of visual SLAM,” in2020 IEEE/RSJ International Conference on Intelligent Robots and Systems (IROS), 2020, pp. 4909–4916, ISSN: 2153-0866

    2020

  32. [32]

    BridgeData v2: A dataset for robot learning at scale,

    H. R. Walke, K. Black, T. Z. Zhao, Q. Vuong, C. Zheng, P. Hansen- Estruch, A. W. He, V . Myers, M. J. Kim, M. Du, A. Lee, K. Fang, C. Finn, and S. Levine, “BridgeData v2: A dataset for robot learning at scale,” inProceedings of The 7th Conference on Robot Learning. PMLR, 2023, pp. 1723–1736, ISSN: 2640-3498

    2023

  33. [33]

    ScanNet: Richly-annotated 3d reconstructions of indoor scenes,

    A. Dai, A. X. Chang, M. Savva, M. Halber, T. Funkhouser, and M. Niessner, “ScanNet: Richly-annotated 3d reconstructions of indoor scenes,” inProceedings of the IEEE Conference on Computer Vision and Pattern Recognition, 2017, pp. 5828–5839

    2017

  34. [34]

    ReFusion: 3d reconstruction in dynamic environments for RGB- d cameras exploiting residuals,

    E. Palazzolo, J. Behley, P. Lottes, P. Giguère, and C. Stachniss, “ReFusion: 3d reconstruction in dynamic environments for RGB- d cameras exploiting residuals,” in2019 IEEE/RSJ International Conference on Intelligent Robots and Systems (IROS), 2019, pp. 7855– 7862, ISSN: 2153-0866

    2019

  35. [35]

    A naturalistic open source movie for optical flow evaluation,

    D. J. Butler, J. Wulff, G. B. Stanley, and M. J. Black, “A naturalistic open source movie for optical flow evaluation,” inComputer Vision – ECCV 2012, A. Fitzgibbon, S. Lazebnik, P. Perona, Y . Sato, and C. Schmid, Eds. Springer, 2012, pp. 611–625

    2012

  36. [36]

    MobileViT: Light-weight, general-purpose, and mobile-friendly vision transformer,

    S. Mehta and M. Rastegari, “MobileViT: Light-weight, general-purpose, and mobile-friendly vision transformer,” inInternational Conference on Learning Representations, 2021

    2021

  37. [37]

    EfficientViT: Lightweight multi-scale attention for high-resolution dense prediction,

    H. Cai, J. Li, M. Hu, C. Gan, and S. Han, “EfficientViT: Lightweight multi-scale attention for high-resolution dense prediction,” inProceed- ings of the IEEE/CVF International Conference on Computer Vision, 2023, pp. 17 302–17 313

    2023