Pith. sign in

REVIEW 5 major objections 7 minor 36 references

RRTrack: Robust and Recoverable Object 6D Pose Tracking for Dynamic Scenes

T0 review · 5 major / 7 minor · reviewed 2026-07-30 · grok-4.5

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

Pith's one-line read RRTrack keeps object 6D pose through fast motion and full disappearance by closing a loop between 2D video segmentation and geometric pose checks, then recovering lost targets with dual template banks at real-time speed.

desk verdict Solid systems paper: geometry-gated VOS↔pose loop plus dual-bank recovery actually moves the needle on fast motion and full occlusion, but the big numbers live almost entirely on their own Isaac Sim benchmark. read the letter →

arxiv 2607.23669 v1 pith:VIJ6KZ6Y submitted 2026-07-26 cs.CV cs.RO

Junyue Li , Ye Zheng , Yifan Chen , Zhe Sun , Xuelong Li This is my paper · ORCID
classification cs.CVcs.RO
keywords object6Dposetrackingrecoverablevideosegmentationrendered-maskagreementtemplatematchingzero-shottraining-freeRGB-Drobotics
verification ladder T0 review T1 audit T2 compute T3 formal

The pith

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

The reading

Robots need continuous 6D object pose, but per-frame estimators are slow and ordinary trackers break when the target moves too fast or vanishes behind an arm, obstacle, or out of view. This paper argues that the fix is not a heavier pose network, but a closed loop: a memory-based 2D segmenter holds the target’s support, a 6D refiner renders a mask from the current pose, and only geometrically consistent observations are allowed to update memory or templates. When agreement collapses or the pose stagnates, a dual-bank matcher—offline synthetic views plus online observation anchors—reinitializes the pose and accepts it only after the same mask check. On a new synthetic RGB-D benchmark of manipulation, agile flight, and factory AGV scenes with 72 loss–reappear events, the method raises equal-subset mean ADD-S accuracy far above a strong local refiner while running above 50 FPS, and qualitative real sequences show the same recovery under noisy handheld depth.

What carries the argument

The 2D–6D closed loop: propagated mask M_t versus pose-rendered mask R_t, scored by precision P_t and support S_t (plus VOS entropy), with EMA–MAD adaptive thresholds that gate snap correction, short/long memory writeback, online bank updates, lost/stagnation detection, and recovery acceptance. Recovery itself is dual-bank DINOv2 CLS retrieval (offline sphere templates + online anchors), refine-and-rank, then accept only if P* clears the tracking threshold.

What would settle it

Run the same equal-subset ADD-S protocol on held-out real RGB-D sequences that have calibrated 6D ground truth and many full occlusions: if closing the rendered-mask loop and dual-bank recovery no longer beat a last-pose refiner on post-reappearance ADD-S AR, or if disabling the agreement gates does not raise drift and recovery delay, the central claim fails.

Watch

Extended reading notes

Core claim

A training-free tracker can survive fast motion and disappearance–reappearance by coupling memory-based video object segmentation with CAD pose refinement under rendered-mask agreement, then recovering with DINOv2 dual-bank template matching. Geometry gates what the 2D memory stores; retrieval plus the same gate restores pose after loss. On the authors’ three-scenario synthetic benchmark this yields large gains in equal-subset mean ADD-S AR and AUC over FoundationPose at 55.2 FPS.

Load-bearing premise

The method assumes that agreement between the 2D mask and the pose-rendered mask, with thresholds learned only on “stable” frames, is enough to stop bad memories and to decide when to snap, declare loss, or accept recovery—across simulation and noisy real depth without needing real 6D ground truth.

Editorial extensions

If this is right

  • Robotic trackers can treat full target loss as a recoverable state rather than a hard failure requiring manual reset.
  • Geometry-gated memory writeback can be reused wherever 2D trackers feed 6D refiners, reducing irreversible drift under occlusion.
  • Offline synthetic templates plus a small online anchor bank can reinitialize novel-object pose without per-object training while staying real-time.
  • Benchmarks for 6D tracking should score lost–reappear events and post-recovery windows, not only smooth visible segments.
  • Symmetric objects remain harder under ADD than ADD-S, so recovery quality will still need symmetry-aware scoring.

Reading between the lines

Editorial extensions of the paper, not claims the author makes directly.

  • The same agreement gate could supervise multi-object scenes by rejecting distractor masks before they enter shared memory.
  • If online anchors dominate under domain shift, the offline bank’s role becomes a safety prior—suggesting adaptive bank weighting as a natural next control knob.
  • Event-level recovery metrics from this benchmark could become a standard stress test for any hybrid 2D–6D tracker claiming long-horizon operation.
Share X Bluesky LinkedIn Reddit HN

Editorial analysis

A structured set of objections, weighed in public.

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

Referee Report

5 major / 7 minor

Summary. The manuscript presents RRTrack, a training-free 6D pose tracker that couples a memory-based VOS module (CUTIE) with a frozen FoundationPose refiner. A rendered-mask agreement mechanism (precision P_t and support S_t between the propagated VOS mask and the pose-rendered mask, Eq. 1) gates memory writeback (Eqs. 5–7), triggers translation "snap" corrections (Eqs. 2–4), declares loss/stagnation (Eqs. 8–9), and validates recovery (Eq. 13). Recovery after disappearance uses DINOv2 CLS-token retrieval over an offline synthetic template bank and an online observation-anchor bank (Eqs. 10–12). The authors also introduce an Isaac Sim RGB-D benchmark (3 scenario families, 22 videos, 38 episodes, 72 lost–reappear events) and report equal-subset mean improvements of 66.3% (ADD-S AR) and 65.7% (ADD-S AUC) over FoundationPose at 55.2 FPS, with qualitative real-world validation on handheld RealSense D435 drone sequences.

Significance. If the results hold, this is a useful practical contribution: the disappearance–reappearance regime is genuinely under-served by existing 6D tracking benchmarks, and the paper ships several things referees should credit explicitly — standard external metrics (ADD/ADD-S at 0.1d), a clearly defined event-level recovery protocol, per-object tables rather than aggregates only, component ablations (Tables III–IV), a credible module-level runtime breakdown (Fig. 4), and a promise to release benchmark and code. The gains on the agile-drone and Dingo subsets are large and consistent (e.g., quad1 ADD-S AR 81.7 vs 9.2 for FoundationPose; Dingo black 71.4 vs 40.8), and the 2D–6D closed-loop design is a sensible engineering response to a real failure mode. The main caveats on significance are that the primary evidence is confined to an author-designed simulator benchmark and that the real-world evidence is qualitative only, so the demonstrated scope is narrower than the abstract's framing suggests.

major comments (5)
  1. [§III.A–B, Eqs. (2)–(13)] Every state transition — snap (Eq. 2), memory writeback (Eqs. 5–7), loss (Eq. 8), stagnation (Eq. 9), and recovery acceptance (Eq. 13) — is gated by thresholds from a single EMA–MAD manager whose statistics are 'updated only on stable tracking frames' (§III.A.b). Stability is itself defined by those same thresholds, so the calibration is self-referential: the paper never states how the manager is initialized before any frame has been classified stable, nor how it behaves when the P_t/S_t distribution shifts (e.g., RealSense depth dropouts and motion blur degrading rendered-mask precision even at correct poses). Table IV ablates adaptive vs. fixed thresholding but provides no sensitivity analysis over α, the 100-frame MAD buffer, or the fixed parameters a_snap, a_lost, θ_th, d_th, W_stag, K_ret, and bank capacities. Since the gating logic — not the retrieval itself — is the paper's coreme
  2. [§IV.B, Table II (GigaPose rows)] GigaPose is described only as 'adapt[ed] to a last pose refiner enabling tracking-efficiency', with no further detail. Its scores collapse to near zero on most rows (e.g., Dingo black ADD-S AR 0.2, agile quad1 9.1, Franka brick 1.8), which is more consistent with a broken tracking adaptation of a single-frame estimator than with the method's known single-frame performance. Because GigaPose is one of three baselines underpinning the state-of-the-art comparison, the adaptation (initialization, hypothesis selection per frame, failure handling) must be documented and sanity-checked — e.g., report its per-frame estimation accuracy on reappearance frames where a correct single-frame estimate should exist — or the baseline should be replaced or removed.
  3. [§IV.B, Table II (MEAN row and Franka block); Abstract] The headline 66.3%/65.7% equal-subset mean gains are driven almost entirely by the agile-drone and Dingo subsets. On Franka-YCB, RRTrack loses clearly to RGBTrack on tuna (ADD-S AUC/AR 41.6/76.9 vs 54.8/99.6) and spam (34.6/56.9 vs 49.8/79.8) and ties on cracker; the equal-subset average over three author-designed scenario families is an author-chosen weighting. The abstract and §I should disclose this subset dependence, and the paper would be substantially strengthened by either (a) quantitative results on at least one pre-existing public RGB-D tracking sequence, or (b) a firm release commitment for the benchmark including the 72 event annotations, so the community can audit the co-designed benchmark/method pair.
  4. [§IV.C, Figs. 5–7; Abstract] The real-world section is explicitly qualitative: two sequences, hand-designed loss events, no annotated reappearance events, no state-transition counts, and no 6D ground truth. Yet the abstract states 'Real-world experiments further validate the robustness of RRTrack under noisy sensing conditions.' Given that the paper's central mechanism is threshold gating whose calibration may shift under real sensing noise (major comment 1), the current evidence cannot support 'validate.' Either add quantitative real-world evidence — e.g., annotated lost/reappear events with recovery success counts and transition statistics, even without 6D ground truth — or downgrade the claim in the abstract and §IV.C to 'qualitatively demonstrate.'
  5. [§IV.A (metrics), Table IV] The event-level recovery criterion (success if any frame in a 10-frame window reaches ADD-S AR ≥ 0.9) is author-chosen and lenient, and the full model succeeds on only 0.722 of events — i.e., 28% of disappearances are never recovered — a figure the abstract omits. Additionally, removing the dual-bank matcher reduces success by only 4.1 points (0.722→0.681) and actually improves median delay (2.0→1.0 frames), which sits awkwardly with the text's claim that each component 'contributes to stable recovery.' Please report sensitivity of success rate to the window length and AR cutoff, state the unrecovered fraction alongside the headline gains, and reconcile the delay inversion for the dual-bank ablation.
minor comments (7)
  1. [§III.A–B] Eqs. (2), (8), (9), and (13) reference τ_P^track, τ_P^floor, and τ_P^stag, but their ordering and relationship are never stated (presumably τ_P^floor < τ_P^track). Note also that Eq. (2) triggers snap below τ_P^track while Eq. (13) accepts recovery above the same threshold; please clarify whether hysteresis is intended.
  2. [Table I] Table I reports videos, episodes, and events but not frame counts or total duration per subset (the 800–1000 frames for agile drone appear only in text). Add these for reproducibility.
  3. [Table II (FPS row)] FoundationPose runs at 98.9 FPS in Table II — faster than RRTrack's 55.2 FPS. The framing 'while achieving 55.2 FPS' should acknowledge that the throughput cost of recovery is relative to a faster tracking-only baseline.
  4. [§III.B.c, final sentence] The 'spherical pose search' fallback for failed retrieval is mentioned once with no description; please give at least a brief specification.
  5. [§I, §III.A] Typos: 'reappearence' (contributions list, §I); 'commited' (§III.A, first paragraph); author block 'Y e Zheng'.
  6. [§IV.B.c] The explanation of the ADD/ADD-S gap via approximate object symmetry (§IV.B.c) is plausible for the quadrotor and Dingo at these resolutions; a one-line per-object symmetry classification (as in BOP) would make this verifiable.
  7. [§IV.A] §IV.A states CUTIE is downsampled to 1/4 resolution and DINOv2 uses 224×224 with 15% context padding — good; please also state the FoundationPose refiner iteration count used per frame, since it affects both accuracy and the 18.1 ms budget.

Circularity Check

0 steps flagged · score 0.0 of 10

No significant circularity: empirical systems paper with external components, standard metrics, and measured gains—not a self-reducing derivation.

full rationale

RRTrack is a training-free engineering pipeline (CUTIE VOS + frozen FoundationPose refiner + DINOv2 dual-bank retrieval) whose central claims are empirical speed/accuracy numbers on a synthetic RGB-D benchmark with simulator GT poses, plus qualitative real-world overlays. ADD/ADD-S AUC and AR are standard external definitions evaluated against independent ground truth; baselines (FoundationPose, GigaPose, RGBTrack) are run on the same streams. Rendered-mask agreement (Eq. 1), snap/lost/stagnation/acceptance gates (Eqs. 2, 5–9, 13), and EMA–MAD thresholds updated on “stable tracking” frames are operational control heuristics, not a claimed first-principles derivation that equates a prediction to its fitted input. Author-chosen recovery windows (10 frames, AR>0.9) and a self-introduced benchmark affect evaluation scope and generalization risk, but do not make the reported improvements true by construction. Load-bearing modules are external (FoundationPose, CUTIE, DINOv2); no uniqueness theorem or ansatz is imported via overlapping-author citation to force the result. No circular step meets the quote-and-reduce bar.

Assumptions & free parameters 6 free parameters · 6 assumptions · 3 invented entities

RRTrack is an engineering composition of frozen models and hand-designed control logic. Load-bearing premises are domain assumptions about RGB-D sensing, CAD availability, and mask–pose agreement as a reliability signal, plus many free thresholds and bank hyperparameters. No new physical entities; invented pieces are system modules (dual banks, snap correction, EMA–MAD manager, synthetic benchmark).

free parameters (6)
  • EMA–MAD adaptive thresholds (τ_P^*, ρ^*, τ_H, floors/stagnation) = α=0.05, MAD buffer=100; metric-specific τ/ρ not fully tabulated
    Tracking, memory writeback, bank update, lost, and recovery acceptance all depend on thresholds derived from running EMA/MAD on self-labeled stable frames; α=0.05 and 100-frame MAD buffer are chosen by hand.
  • Snap/lost mask-area thresholds a_snap, a_lost
    Minimum propagated-mask area gates local translation snap vs declaring loss; values are design choices affecting recovery frequency.
  • Stagnation kinematic thresholds θ_th, d_th and window W_stag
    Fixed pose-change thresholds decide optimizer stagnation and force recovery.
  • Dual-bank retrieval K_ret and bank capacities = K_ret=3, N_off=256, N_on≤64
    Top-K=3 per bank, offline 256 views with 180° in-plane aug, online 64 FIFO anchors directly affect recovery recall/latency.
  • DINOv2 input recipe (224×224, 15% context padding) = 224×224, 15% padding
    Descriptor extraction settings chosen for recovery matching; not learned end-to-end but still free design knobs.
  • Recovery success criterion (10-frame window, ADD-S AR≥0.9) = 10 frames, AR>0.9
    Event-level success and post-recovery scores depend on this evaluation definition.
assumptions (6)
  • domain assumption A known object CAD/mesh is available at test time for rendering and FoundationPose refinement.
    Model-based zero-shot setting stated throughout; tracking and recovery render masks/poses from the mesh.
  • domain assumption Synchronized RGB-D and known intrinsics K suffice to back-project mask centers and refine 6D pose.
    Snap correction and refiner inputs assume usable depth inside the mask (Eq. 3–4).
  • ad hoc to paper Agreement between VOS mask M_t and pose-rendered mask R_t is a valid proxy for geometric tracking reliability and safe memory writeback.
    Core closed-loop design in §III.A; gates I_long_mem, I_short_mem, I_bank, I_lost, I_acc all rest on P_t/S_t.
  • domain assumption Frozen FoundationPose refiner/ranker and CUTIE VOS behave as reliable black boxes under the paper’s motion and occlusion regime when gated as described.
    Training-free pipeline; no fine-tuning; performance inherits these models’ failure modes.
  • domain assumption DINOv2 CLS tokens on masked crops are discriminative enough to retrieve coarse pose hypotheses across synthetic↔real and pre/post-occlusion appearance change.
    Dual-bank recovery §III.B.b; acceptance still needs refiner + mask check, but retrieval quality is assumed adequate.
  • ad hoc to paper Isaac Sim scenarios (Franka-YCB, agile quad, Dingo) with equal-subset averaging represent the target dynamic robotic tracking regime for claimed gains.
    Primary quantitative claims are on this author-built benchmark (§IV.B).
invented entities (3)
  • Rendered-mask agreement closed loop (P_t/S_t-gated VOS↔pose memory)
    purpose: Verify 2D support with 6D geometry before memory updates and trigger snap/recovery.
    System-level mechanism, not a physical entity; evidence is ablations and tracking metrics on the new benchmark.
  • DINOv2 dual-bank recovery (offline synthetic templates + online observation anchors)
    purpose: Re-initialize pose after disappearance–reappearance without manual reset.
    Architectural construct; supported by recovery-event ablation, not external independent measurement.
  • RRTrack synthetic RGB-D benchmark (3 scenario families, 72 lost–reappear events)
    purpose: Evaluate fast motion, full occlusion, and recovery in robotic settings underrepresented in BOP-style sets.
    New dataset asset; valuable if released, but currently author-controlled evaluation substrate.

how reviews work

0 comments
Cite this review

Pith. "Pith review of RRTrack: Robust and Recoverable Object 6D Pose Tracking for Dynamic Scenes." pith.science (2026). https://pith.science/paper/VIJ6KZ6Y

@misc{pith2026260723669,
  author       = {Pith},
  title        = {Pith review of: RRTrack: Robust and Recoverable Object 6D Pose Tracking for Dynamic Scenes},
  year         = {2026},
  howpublished = {\url{https://pith.science/paper/VIJ6KZ6Y}},
  note         = {Machine review of arXiv:2607.23669}
}
read the original abstract

Robust object 6D pose tracking is critical for robotic systems operating in dynamic and occluded scenes. Per-frame estimators are accurate but computationally expensive, while current trackers struggle with fast motion and complete occlusion due to their reliance on continuous visibility. To address these challenges, we present RRTrack, an efficient, recoverable object 6D pose tracker that enables robust tracking through fast motion and target disappearance--reappearance. RRTrack introduces a 2D--6D closed-loop tracking strategy that integrates memory-based video object segmentation (VOS) with 6D pose refinement. The 2D branch maintains target localization, and the 6D branch verifies geometric consistency before memory updates. In addition, a DINOv2-based dual-bank template matching module is developed to recover lost targets by jointly exploiting offline synthetic templates and online observation anchors while maintaining real-time efficiency. We also introduce a synthetic RGB-D benchmark comprising three robotic scenarios with fast motion and full occlusion. Experimental results on the synthetic benchmark demonstrate that RRTrack improves equal-subset mean ADD-S AR by 66.3\% and ADD-S AUC by 65.7\% over FoundationPose while achieving 55.2 FPS. Real-world experiments further validate the robustness of RRTrack under noisy sensing conditions. Project page: https://github.com/7kevin24/RRTrack

Figures

Figures reproduced from arXiv: 2607.23669 by the authors.

Figure 1
Figure 1. Motivation and performance overview of the proposed RRTrack. Top: the target [PITH_FULL_IMAGE:figures/full_fig_p001_1.png] view at source ↗
Figure 2
Figure 2. Overview of RRTrack. Given an RGB–D stream, VOS propagates the object mask [PITH_FULL_IMAGE:figures/full_fig_p003_2.png] view at source ↗
Figure 3
Figure 3. Demonstration of the proposed synthetic benchmark. Each column presents three [PITH_FULL_IMAGE:figures/full_fig_p006_3.png] view at source ↗
Figures from the paper (4 more)
Figure 4
Figure 4. Figure 4: Average runtime of different modules on the synthetic benchmarks. Fine-grained [PITH_FULL_IMAGE:figures/full_fig_p006_4.png]
Figure 5
Figure 5. Figure 5: Real-world experimental setups. The upper panel shows RGB-D capture using [PITH_FULL_IMAGE:figures/full_fig_p007_5.png]
Figure 6
Figure 6. Figure 6: Temporal qualitative comparison between FoundationPose and RRTrack on consecutive frames from the underground parking-garage setting. FoundationPose is initially [PITH_FULL_IMAGE:figures/full_fig_p008_6.png]
Figure 7
Figure 7. Figure 7: Real-world disappearance–reappearance examples. Each row shows one event: at [PITH_FULL_IMAGE:figures/full_fig_p008_7.png]

Discussion (0). Continue with ORCID to comment.

Reference graph

Works this paper leans on

36 extracted references · 2 canonical work pages

  1. [1]

    A practical robotic grasping method by using 6-D pose estimation with protective correction,

    H. Zhang, Z. Liang, C. Li, H. Zhong, L. Liu, C. Zhao, Y . Wang, and Q. M. J. Wu, “A practical robotic grasping method by using 6-D pose estimation with protective correction,”IEEE Transactions on Industrial Electronics, vol. 69, DOI 10.1109/TIE.2021.3075836, no. 4, pp. 3876– 3886, 2022

    arXiv 2021

  2. [2]

    Real- time markerless tracking for augmented reality: The virtual visual servoing framework,

    A. I. Comport, E. Marchand, M. Pressigout, and F. Chaumette, “Real- time markerless tracking for augmented reality: The virtual visual servoing framework,”IEEE Transactions on Visualization and Computer Graphics, vol. 12, no. 4, pp. 615–628, 2006

    2006

  3. [3]

    Socially conscious navigation of mobile robots based on deep reinforcement learning,

    Y . Kong, X. Gong, Y . Wang, J. Yu, B. Lu, W. Chi, and L. Sun, “Socially conscious navigation of mobile robots based on deep reinforcement learning,”IEEE Transactions on Industrial Electronics, vol. 72, DOI 10.1109/TIE.2025.3566742, no. 12, pp. 13 542–13 551, 2025

    arXiv 2025

  4. [4]

    Deep learning-based object pose estimation: A comprehensive survey,

    J. Liu, W. Sun, H. Yang, Z. Zeng, C. Liu, J. Zheng, X. Liu, H. Rahmani, N. Sebe, and A. Mian, “Deep learning-based object pose estimation: A comprehensive survey,”International Journal of Computer Vision, vol. 134, DOI https://doi.org/10.1007/s11263-025-02646-6, no. 81, pp. 1–45, 2026

    doi:10.1007/s11263-025-02646-6 2026

  5. [5]

    FoundationPose: Unified 6D pose estimation and tracking of novel objects,

    B. Wen, W. Yang, J. Kautz, and S. Birchfield, “FoundationPose: Unified 6D pose estimation and tracking of novel objects,” inProceedings of the IEEE/CVF International Conference on Computer Vision (CVPR), pp. 17 868–17 879, 2024

    2024

  6. [6]

    GigaPose: Fast and robust novel object pose estimation via one correspondence,

    V . N. Nguyen, T. Groueix, M. Salzmann, and V . Lepetit, “GigaPose: Fast and robust novel object pose estimation via one correspondence,” inProceedings of the IEEE/CVF International Conference on Computer Vision (CVPR), pp. 9903–9913, 2024

    2024

  7. [8]

    Dynamicpose: Real-time and robust 6d object pose tracking for fast-moving cameras and objects,

    T. Liang, Y . Zeng, J. Xie, and B. Zhou, “Dynamicpose: Real-time and robust 6d object pose tracking for fast-moving cameras and objects,” in2025 IEEE/RSJ International Conference on Intelligent Robots and Systems (IROS), DOI 10.1109/IROS60139.2025.11247403, pp. 2424– 2431, 2025

    arXiv 2025

  8. [9]

    BOP: Benchmark for 6D object pose estimation,

    T. Hodan, F. Michel, E. Brachmann, W. Kehl, A. GlentBuch, D. Kraft, B. Drost, J. Vidal, S. Ihrke, X. Zabuliset al., “BOP: Benchmark for 6D object pose estimation,” inProceedings of the European Conference on Computer Vision (ECCV), pp. 19–34, 2018

    2018

Show all 36 references
  1. [10]

    Category-level 6-D ob- ject pose estimation with learnable prior embeddings for robotic grasping,

    S. Yu, D.-H. Zhai, J. Yin, and Y . Xia, “Category-level 6-D ob- ject pose estimation with learnable prior embeddings for robotic grasping,”IEEE Transactions on Industrial Electronics, vol. 72, DOI 10.1109/TIE.2025.3555019, no. 11, pp. 11 682–11 694, 2025

    2025

  2. [11]

    Zephyr: Zero-shot pose hypothesis rating,

    B. Okorn, Q. Gu, M. Hebert, and D. Held, “Zephyr: Zero-shot pose hypothesis rating,” inProceedings of the IEEE International Conference on Robotics and Automation (ICRA), pp. 14 141–14 148. IEEE, 2021

    2021

  3. [12]

    OVE6D: Object viewpoint en- coding for depth-based 6D object pose estimation,

    D. Cai, J. Heikkil ¨a, and E. Rahtu, “OVE6D: Object viewpoint en- coding for depth-based 6D object pose estimation,” inProceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recogni- tion(CVPR), pp. 6803–6813, 2022

    2022

  4. [13]

    MegaPose: 6D Pose Estimation of Novel Objects via Render & Compare,

    Y . Labb ´e, L. Manuelli, A. Mousavian, S. Tyree, S. Birchfield, J. Trem- blay, J. Carpentier, M. Aubry, D. Fox, and J. Sivic, “MegaPose: 6D Pose Estimation of Novel Objects via Render & Compare,” inCoRL, 2022

    2022

  5. [14]

    Genflow: Generalizable recurrent flow for 6d pose refinement of novel objects,

    S. Moon, H. Son, D. Hur, and S. Kim, “Genflow: Generalizable recurrent flow for 6d pose refinement of novel objects,” inProceedings of the IEEE/CVF International Conference on Computer Vision (CVPR), pp. 10 039–10 049, 2024

    2024

  6. [15]

    Sam-6d: Segment anything model meets zero-shot 6d object pose estimation,

    J. Lin, L. Liu, D. Lu, and K. Jia, “Sam-6d: Segment anything model meets zero-shot 6d object pose estimation,” inProceedings of the IEEE/CVF International Conference on Computer Vision (CVPR), pp. 27 906–27 916, Jun. 2024

    2024

  7. [16]

    Keypoint- guided efficient pose estimation and domain adaptation for mi- cro aerial vehicles,

    Y . Zheng, C. Zheng, J. Shen, P. Liu, and S. Zhao, “Keypoint- guided efficient pose estimation and domain adaptation for mi- cro aerial vehicles,”IEEE Transactions on Robotics, vol. 40, DOI 10.1109/TRO.2024.3400938, pp. 2967–2983, 2024

    2024

  8. [17]

    ZS6D: Zero-shot 6D object pose estimation using vision transformers,

    P. Ausserlechner, D. Haberger, S. Thalhammer, J.-B. Weibel, and M. Vincze, “ZS6D: Zero-shot 6D object pose estimation using vision transformers,” inProceedings of the IEEE International Conference on Robotics and Automation (ICRA), pp. 463–469. IEEE, 2024

    2024

  9. [18]

    Foundpose: Unseen object pose estimation with foundation features,

    E. P. ¨Ornek, Y . Labb´e, B. Tekin, L. Ma, C. Keskin, C. Forster, and T. Hodaˇn, “Foundpose: Unseen object pose estimation with foundation features,” inProceedings of the IEEE/CVF International Conference on Computer Vision (ECCV), 2024

    2024

  10. [19]

    Cnos: A strong baseline for cad-based novel object segmentation,

    V . N. Nguyen, T. Groueix, G. Ponimatkin, V . Lepetit, and T. Hodan, “Cnos: A strong baseline for cad-based novel object segmentation,” in Proceedings of the IEEE/CVF International Conference on Computer Vision Workshop (ICCVW), pp. 2134–2140, 2023

    2023

  11. [20]

    Segment anything,

    A. Kirillov, E. Mintun, N. Ravi, H. Mao, C. Rolland, L. Gustafson, T. Xiao, S. Whitehead, A. C. Berg, W.-Y . Lo, P. Doll ´ar, and R. Girshick, “Segment anything,” 2023. [Online]. Available: https: //arxiv.org/abs/2304.02643

    2023 arXiv

  12. [21]

    Freeze: Training- free zero-shot 6d pose estimation with geometric and vision foundation models,

    A. Caraffa, D. Boscaini, A. Hamza, and F. Poiesi, “Freeze: Training- free zero-shot 6d pose estimation with geometric and vision foundation models,” inProceedings of the IEEE/CVF International Conference on Computer Vision (ECCV), 2024

    2024

  13. [22]

    Learning general and distinctive 3D local deep descriptors for point cloud registration,

    F. Poiesi and D. Boscaini, “Learning general and distinctive 3D local deep descriptors for point cloud registration,”IEEE Transactions on Pattern Analysis and Machine Intelligence, 2023

    2023

  14. [23]

    Object pose estimation via the aggregation of diffusion features,

    T. Wang, G. Hu, and H. Wang, “Object pose estimation via the aggregation of diffusion features,” inProceedings of the IEEE/CVF International Conference on Computer Vision (CVPR), 2024

    2024

  15. [24]

    Diffusion Features for Zero-Shot 6DoF Object Pose Estimation,

    B. V on Gimborn, P. Ausserlechner, M. Vincze, and S. Thalhammer, “Diffusion Features for Zero-Shot 6DoF Object Pose Estimation,” in Proceedings of the IEEE/CVF International Conference on Computer Vision (CVPR), 2024

    2024

  16. [25]

    DINOv2: Learning robust visual features without supervision,

    M. Oquab, T. Darcet, T. Moutakanni, H. V o, M. Szafraniec, V . Khalidov, P. Fernandez, D. Haziza, F. Massa, A. El-Noubyet al., “DINOv2: Learning robust visual features without supervision,”arXiv preprint arXiv:2304.07193, 2023

    2023 arXiv

  17. [26]

    Ai flow: Perspectives, scenarios, and approaches,

    “Ai flow: Perspectives, scenarios, and approaches,”AI and Ethics, DOI 10.1007/s44336-025-00031-y, 2026. [Online]. Available: https: //doi.org/10.1007/s44336-025-00031-y

    2026 doi

  18. [27]

    Ai flow at the network edge,

    “Ai flow at the network edge,”IEEE Communications Magazine, 2025. [Online]. Available: https://ieeexplore.ieee.org/document/10884554

    2025

  19. [28]

    Generative transmission: Rethinking computation, bandwidth, and memory in communication,

    X. Chen, J. Luo, Y . Fan, H. Huang, C. Zhang, and X. Li, “Generative transmission: Rethinking computation, bandwidth, and memory in communication,”arXiv preprint arXiv:2607.17482, DOI 10.48550/arXiv.2607.17482, 2026. [Online]. Available: https://arxiv.org/ abs/2607.17482

    2026 doi arXiv

  20. [29]

    Optical image processing and applications empowered by vision-language models,

    J. Xiao, Z. Sun, H. An, H. Zhao, and X. Li, “Optical image processing and applications empowered by vision-language models,”iOptics, vol. 1, no. 1, p. 100003, 2025

    2025

  21. [30]

    DeepIM: Deep iterative matching for 6D pose estimation,

    Y . Li, G. Wang, X. Ji, Y . Xiang, and D. Fox, “DeepIM: Deep iterative matching for 6D pose estimation,” inProceedings of the European Conference on Computer Vision (ECCV), pp. 683–698, 2018

    2018

  22. [31]

    Poserbpf: A rao-blackwellized particle filter for 6d ob- ject pose tracking,

    X. Deng, Y . Xiang, A. Mousavian, C. Eppner, T. Bretl, and D. Fox, “Poserbpf: A rao-blackwellized particle filter for 6d ob- ject pose tracking,”IEEE Transactions on Robotics, vol. 37, DOI 10.1109/TRO.2021.3056043, no. 5, pp. 1328–1342, 2021

    2021

  23. [32]

    Iterative corresponding geometry: Fusing region and depth for highly efficient 3D tracking of textureless objects,

    M. Stoiber, M. Sundermeyer, and R. Triebel, “Iterative corresponding geometry: Fusing region and depth for highly efficient 3D tracking of textureless objects,” inProceedings of the IEEE/CVF International Conference on Computer Vision (CVPR), pp. 6855–6865, 2022

    2022

  24. [33]

    BundleTrack: 6D pose tracking for novel objects without instance or category-level 3D models,

    B. Wen and K. Bekris, “BundleTrack: 6D pose tracking for novel objects without instance or category-level 3D models,” in2021 IEEE/RSJ International Conference on Intelligent Robots and Systems (IROS), pp. 8067–8074. IEEE, 2021

    2021

  25. [34]

    Putting the Object Back into Video Object Segmentation,

    H. K. Cheng, S. W. Oh, B. Price, J.-Y . Lee, and A. G. Schwing, “Putting the Object Back into Video Object Segmentation,” inProceedings of the IEEE/CVF International Conference on Computer Vision (CVPR), 2024

    2024

  26. [35]

    PoseCNN: A Con- volutional Neural Network for 6D Object Pose Estimation in Cluttered Scenes,

    Y . Xiang, T. Schmidt, V . Narayanan, and D. Fox, “PoseCNN: A Con- volutional Neural Network for 6D Object Pose Estimation in Cluttered Scenes,” inRobotics: Science and Systems (RSS), 2018

    2018

  27. [36]

    SAM 2: Segment anything in images and videos,

    N. Ravi, V . Gabeur, Y .-T. Hu, R. Hu, C. Ryali, T. Ma, H. Khedr, R. R ¨adle, C. Rolland, L. Gustafsonet al., “SAM 2: Segment anything in images and videos,”arXiv preprint arXiv:2408.00714, 2024

    2024 arXiv

  28. [37]

    Joint Feature Learning and Relation Modeling for Tracking: A One-Stream Framework,

    B. Ye, H. Chang, B. Ma, S. Shan, and X. Chen, “Joint Feature Learning and Relation Modeling for Tracking: A One-Stream Framework,” in Proceedings of the European Conference on Computer Vision (ECCV), 2022

    2022

Pith tools

Reviewed July 30, 2026 · model on record in the stance chip above.