Pith. sign in

REVIEW 3 major objections 5 minor 43 references

OSSDD - a New Open Dataset for Sentinel-1 Ship Detection

T0 review · 3 major / 5 minor · reviewed 2026-08-04 · deepseek-v4-flash

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

Pith's one-line read This paper introduces OSSDD, a freely available Sentinel-1 ship detection dataset with 15,197 float-format VV/VH patches and 55,759 ship annotations, and shows that standard detectors can be trained on it, with Faster R-CNN reaching 0.70 mA

desk verdict OSSDD provides a valuable float-format SAR ship detection dataset, but its benchmark numbers are weakened by an underspecified and non-independent annotation pipeline and by internal count inconsistencies that must be fixed. read the letter →

arxiv 2608.01963 v1 pith:YSNET4KH submitted 2026-08-03 cs.CV

Horst Hammer , Sylvia Hochstuhl , Antje Thiele , Tobias Brosch , Padraig Davidson , Tim Remiger , Michael Teutsch This is my paper
classification cs.CV
keywords SARshipdetectionSentinel-1OpenSARShipdatasetobjectrotatedboundingboxessyntheticapertureradar
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

The paper's central claim is that OSSDD is a valid, usable ship detection dataset: it contains 15,197 Sentinel-1 Ground-Range Detected amplitude patches in VV and VH polarization, with binary masks, axis-aligned bounding boxes, and rotated bounding boxes for 55,759 ships, all stored in 32-bit float format rather than the 8-bit or JPEG formats common in existing SAR ship datasets. The authors argue that float format removes a key obstacle, because models trained on quantized or lossy images may degrade when applied to native SAR amplitude data. To demonstrate usability, they train three common detectors (Faster R-CNN, FCOS, DETR) on OSSDD; the best result is Faster R-CNN at 0.70 mAP on the VH test set. If correct, the dataset gives the community a reproducible free benchmark for training and comparing SAR ship detectors.

What carries the argument

The central object is OSSDD itself: a float-format, dual-polarization dataset of Sentinel-1 amplitude patches. The supporting mechanism that carries the argument is the semi-automatic VH-threshold annotation pipeline. VH polarization gives strong ship-water contrast, so a threshold plus connected-component extraction yields candidate ship masks; manual post-processing removes land clutter, merged small ships, and bright-scatterer signatures; the remaining components define the binary masks and both bounding box formats. This pipeline is what turns scattered OpenSARShip classification chips into coherent detection ground truth, and the float patches preserve the native amplitude range that op

What would settle it

Independently re-annotate a random sample of, say, 50 test patches with SAR experts, then compare ship counts and box coordinates to the OSSDD labels; if many original masks are missing ships that experts confidently identify, or if the exclusion rules remove vessels that experts would label, the annotation premise fails and the mAP numbers cannot be taken at face value.

Watch

Extended reading notes

Core claim

The discovery is the dataset itself and its construction. Starting from the source scenes of OpenSARShip, the authors extract 700x700-pixel patches around known ships and re-annotate them through a semi-automatic pipeline: a heuristic threshold on VH-polarized amplitude, connected-component extraction, removal of very small, narrow, or bright-scatterer components, manual cleanup, and computation of binary masks plus axis-aligned and rotated bounding boxes. The result is 15,197 patches with 55,759 ship annotations split into train, validation, and test sets, with test and validation patches center-cropped to 512x512. Training experiments show that standard detectors learn from the data, with

Load-bearing premise

The load-bearing premise is that the semi-automatic annotation pipeline in Section 4.1—an unreported heuristic threshold on VH images, connected-component extraction, and manual cleanup—produces ground-truth labels accurate and complete enough for training and evaluating detectors; if those labels are wrong or incomplete, the benchmark numbers and the claim that OSSDD is well-annotated do not stand.

Editorial extensions

If this is right

  • Freely available 32-bit float patches in VV and VH let researchers train and evaluate detectors on data consistent with native Sentinel-1 GRD amplitude, avoiding unknown 8-bit quantization steps.
  • The reported Faster R-CNN results (0.70 mAP on VH, 0.62 mAP on VV) provide concrete baselines that future detectors can be compared against on the same splits.
  • Because annotations include binary masks, axis-aligned boxes, and rotated boxes, the same dataset supports segmentation, standard detection, and rotated detection architectures.
  • The train/validation/test split keeps patches from the same Sentinel-1 scene in one split, reducing the chance that benchmark scores are inflated by near-duplicate scene content.
  • Land-containing patches are balanced across splits, giving the test set a more realistic mix of open-water and coastal scenarios.

Reading between the lines

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

  • Because ships with very bright scatterers or weak signatures were deliberately excluded from ground truth, detectors that find them are scored as false positives; the reported mAP may understate real-world detection ability rather than overstate it.
  • All patches come from the same 41 scenes centred on five large harbours, so the benchmark alone says little about open-ocean or geographically different conditions; a cross-region evaluation is needed to test transfer.
  • The threshold and connected-component rules of the annotation pipeline are described qualitatively, not specified numerically; publishing those parameters or releasing intermediate binary masks would let others reproduce the labels exactly.
  • A useful stress test would be training on OSSDD and evaluating on recent Sentinel-1 scenes over different ports or at sea; strong retention of mAP would support the paper's claim that float data improves transfer to unseen SAR imagery.
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

3 major / 5 minor

Summary. The paper introduces OSSDD, a new Sentinel-1 ship detection dataset derived from OpenSARShip 1.0. The dataset contains 15,197 image patches in VV and VH polarization, provided as 32-bit float TIF files, together with binary ship masks, axis-aligned bounding boxes, and rotated bounding boxes for a total of 55,759 ships. The authors describe a semi-automatic annotation pipeline built on a heuristic VH threshold, connected-component extraction, manual post-processing, and cropping of validation/test patches. They train Faster R-CNN, FCOS, and DETR on the training split and report benchmark results on the test split, with Faster R-CNN achieving the best mAP of 0.70 on VH. The dataset is available on Hugging Face.

Significance. If the dataset is sound, it makes a useful contribution: unlike most existing SAR ship detection datasets, OSSDD distributes 32-bit float amplitudes rather than 8-bit quantized images, which addresses a real concern about transferability to unseen Sentinel-1 data. The inclusion of both VV/VH polarizations, binary masks, and rotated bounding boxes broadens the range of usable detector architectures. The split-by-scene protocol and the explicit discussion of difficult cases are also positive features. However, the central claim that OSSDD is a well-annotated, reliable training and evaluation resource rests on the unvalidated and only partially specified annotation pipeline of Section 4.1. The paper should be credited for releasing the data openly and for providing baseline experiments, but the reported benchmark numbers cannot be taken as evidence of detection performance on all ships present until the annotation process is made reproducible and independently validated.

major comments (3)
  1. [Section 4.1 and Fig. 5] The annotation pipeline is load-bearing but not reproducible. The 'heuristic threshold' on VH images is not specified, connected-component removal rules are described only as 'very small or very narrow (1 pixel wide)', and manual post-processing criteria are absent. No annotation code or tool is released. More importantly, the paper explicitly excludes actual ships from the ground truth: bright-scatterer ships, weak-signature ships, and merged small ships are omitted (Section 4.1, Fig. 5). Detectors that find such ships are then counted as false positives in the benchmark (Fig. 5). Consequently, the mAP values in Tables 6 and 7 measure agreement with a curated subset, not detection performance on all ships present. Please release the threshold/parameters and annotation code, quantify the number and nature of excluded ships, and validate labels against independent re-annotation or AIS dat
  2. [Section 5 and Table 3] There are unresolved count inconsistencies that undermine confidence in the metadata. The text states that 15,161 raw patches were extracted, but Table 3's total is 15,197 and the split sizes sum to 15,197; cropping to 512x512 for validation/test should not increase the number of patches. For AABB objects, 37,236 + 15,573 + 2,990 = 55,799, not the stated 55,759. The validation patch text says 2,290 annotated ships for binary masks/AABBs, while Table 3 lists 2,990. These discrepancies need to be reconciled or explicitly explained; as written they cast doubt on the internal consistency of the released metadata.
  3. [Section 6.3] The evaluation protocol is under-specified. Precision/recall are computed at IoU 0.2 and confidence 0.3, and mAP is averaged over IoU thresholds from 0.1 to 0.95 in steps of 0.05, but the matching rule between detections and ground truth is not stated. In particular, it is unclear how duplicate detections, detections on land, or detections of ships that were excluded from the ground truth are handled. Given that the ground truth intentionally omits some true ships and land structures, a precise matching protocol or release of evaluation code is necessary for the benchmark numbers to be interpretable and reproducible.
minor comments (5)
  1. [Section 6.1] Faster R-CNN is described as a 'two-stage anchor-free detector'; it is anchor-based. This should be corrected to avoid confusion about the architectures.
  2. [Abstract] The dataset name is typeset inconsistently as 'OpenSARShip-ShipDetectionDataset' without spaces; use 'OpenSARShip-Ship Detection Dataset' consistently.
  3. [Section 5] The paper notes that RBBs with vertices outside the cropped area are discarded, making RBB counts differ from AABB/mask counts. This is legitimate, but it should be clearly documented in the dataset card and metadata.csv, including the exact numbers per split.
  4. [Table 1] For OSSDD, the 'Objects' column lists 55,759, but this number is not directly comparable to other datasets because it excludes ships dropped during cropping in the RBB annotations; consider clarifying the counting convention in the table caption.
  5. [General] The paper does not specify a license for the dataset. Since the data are derived from Sentinel-1/Copernicus data and includes OpenSARShip annotations, a license and attribution statement would be useful for users.

Circularity Check

1 steps flagged · score 2.0 of 10

No derivation is circular; a mild self-referential evaluation loop affects how mAP can be interpreted as evidence of annotation quality.

  1. self definitional [Section 4.1 and Figure 5 / Section 7]
    "The ship located at the right side of the image showing a very bright scattering center was excluded from the ground truth data because the very bright scatterer would lead to very large bounding boxes which may mislead the training. Yet, all three detectors were able to find the ship and provide it with a reasonable bounding box. This will however be classified as a misdetection."

    The ground truth is constructed by a heuristic threshold plus connected-component filtering and manual exclusion of bright-scatterer ships. Any detector output that locates such an excluded real ship is, by construction, judged a false positive, because the label set was defined to omit it. Thus mAP/precision/recall measure agreement with the annotation pipeline's inclusion decisions, not performance against an independent definition of 'all ships present.' The Section 7 claim that mAP=0.70 'demonstrat[es] that OSSDD provides a sufficiently rich and well-annotated training basis' therefore uses as evidence a metric whose reference labels were produced by the same curation step. This is a mild self-referential evaluation loop, not a derivation of a first-principles result.

full rationale

The paper's central contribution is an empirical dataset and benchmark measurements, not a first-principles derivation. No fitted parameter is renamed as a prediction, no self-citation chain carries the argument, and no uniqueness theorem is imported. The dataset is built from OpenSARShip metadata plus a semi-automatic annotation pipeline; the detector results are simply measurements on a held-out subset. The only notable circularity-adjacent aspect is that the annotations themselves are the reference for evaluating detectors, and the annotation policy explicitly excludes some real ships (bright-scatterer ships, small close ships, weak-signature ships). Detections of those excluded ships are counted as false positives by construction, so the reported mAP reflects agreement with the pipeline's conventions as much as detection quality. This is disclosed in the paper and weakens the strength of the 'well-annotated' conclusion, but it does not reduce the dataset construction or benchmark numbers to their own inputs. Score 2 reflects this mild self-referential evaluation loop, not a structurally circular derivation.

Assumptions & free parameters 3 free parameters · 4 assumptions · 0 invented entities

The central claim rests on the correctness of the OpenSARShip-derived seeds and the threshold-based annotation pipeline. The pipeline's heuristic threshold, manual exclusion rules, and land/water split are all assumptions about the quality of the resulting labels. No independent validation is provided.

free parameters (3)
  • VH heuristic threshold for ship mask binarization = not reported
    Introduced in Section 4.1 to turn VH amplitude into a binary ship mask; its value is not stated, so the annotations are not exactly reproducible.
  • Component size and narrowness removal rules = not reported
    Very small and 1-pixel-wide components are discarded in Section 4.1; these cutoffs are hand-set and change which ships appear in the ground truth.
  • IoU and confidence thresholds for precision/recall = IoU 0.2, confidence 0.3
    Hand-chosen evaluation thresholds in Section 6.3; they affect the reported precision and recall and are not justified.
assumptions (4)
  • domain assumption OpenSARShip annotations correctly identify real ships in the Sentinel-1 scenes
    Used as seeds for patch extraction and, for old GRD data, as part of the ground truth (Sections 3-4). No independent verification is reported.
  • domain assumption VH polarization contrast is sufficient for threshold-based ship extraction
    The entire annotation pipeline in Section 4.1 depends on this; no quantitative analysis of missed or false components is given.
  • domain assumption The GRD products downloaded from Copernicus correspond to the same scenes as OpenSARShip and, for SLC images, can be re-annotated independently
    Section 4.2 relies on this correspondence; the paper notes annotations were incompatible for SLC and had to be redone.
  • domain assumption SNAP fractional water mask correctly separates land and water for split balancing
    Used in Section 5 to balance land-containing patches across train/test/validation splits.

how reviews work

0 comments
Cite this review

Pith. "Pith review of OSSDD - a New Open Dataset for Sentinel-1 Ship Detection." pith.science (2026). https://pith.science/paper/YSNET4KH

@misc{pith2026260801963,
  author       = {Pith},
  title        = {Pith review of: OSSDD - a New Open Dataset for Sentinel-1 Ship Detection},
  year         = {2026},
  howpublished = {\url{https://pith.science/paper/YSNET4KH}},
  note         = {Machine review of arXiv:2608.01963}
}
read the original abstract

Ship detection in Synthetic Aperture Radar (SAR) images plays an important role for maritime situational awareness, especially with respect to different illegal activities at sea such as illegal fishing, smuggling or border violations. Modern ship detection methods using neural networks usually require large training datasets, which are considerably scarcer in the SAR domain than in the electro-optical domain. While several free datasets exist for this task, their availability and usability vary. In this paper, OpenSARShip-Ship Detection Dataset (OSSDD), a new dataset based on the well-known OpenSARShip 1.0 dataset is proposed for training neural networks for SAR ship detection. OSSDD is freely available and contains 15,197 Sentinel-1 amplitude patches in VV and VH polarization, binary ship masks, axis-aligned bounding box and rotated bounding box annotations for a total of 55,759 ships. The construction of the dataset, the contents and structure of the downloadable data and experiments with three common detector models (Faster R-CNN, FCOS, DETR) are shown and discussed. The results serve as benchmarks for future experiments. The dataset is available on Hugging Face at https://huggingface.co/datasets/sylviaHoch/OpenSARShip-Ship-Detection-Dataset.

Figures

Figures reproduced from arXiv: 2608.01963 by the authors.

Figure 1
Figure 1. Sample scenes with annotated (green boxes) and missing (red circles) ships in OpenSARShip. [PITH_FULL_IMAGE:figures/full_fig_p004_1.png] view at source ↗
Figure 2
Figure 2. Examples of ship mask extractions: Left column: VH polarized Sentinel-1 scene; middle column: VV polarized [PITH_FULL_IMAGE:figures/full_fig_p005_2.png] view at source ↗
Figure 3
Figure 3. Detection results for ships on open water. a) Result of Faster R-CNN; b) Result of FCOS; c) Result of DETR; [PITH_FULL_IMAGE:figures/full_fig_p010_3.png] view at source ↗
Figures from the paper (2 more)
Figure 4
Figure 4. Figure 4: Detection results for smaller ships close together and close to land structures. a) Result of Faster R-CNN; b) [PITH_FULL_IMAGE:figures/full_fig_p010_4.png]
Figure 5
Figure 5. Figure 5: Detection results for difficult ships. a) Result of Faster R-CNN; b) Result of FCOS; c) Result of DETR; d) Ground truth [PITH_FULL_IMAGE:figures/full_fig_p011_5.png]

Discussion (0). Sign in to comment.

Reference graph

Works this paper leans on

43 extracted references · 3 canonical work pages

  1. [1]

    Mahony and Rodolphe Sepulchre , title =

    Pierre-Antoine Absil and Robert E. Mahony and Rodolphe Sepulchre , title =

  2. [2]

    , author =

    Sur quelques propriétés des déterminants gauches. , author =. Journal für die reine und angewandte Mathematik , doi =. 1846 , lastchecked =

  3. [3]

    Electronic Journal of Qualitative Theory of Differential Equations , number = 13, volume =

    A. Electronic Journal of Qualitative Theory of Differential Equations , number = 13, volume =

  4. [4]

    The American Mathematical Monthly , volume = 53, number = 3, month = 3, doi =

    Harold Scott MacDonald Coxeter , year = 1946, title =. The American Mathematical Monthly , volume = 53, number = 3, month = 3, doi =

    1946

  5. [5]

    , year = 1995, title =

    Fisher, Nicholas I. , year = 1995, title =

    1995

  6. [6]

    Photogrammetrie -- Fernerkundung -- Geoinformatik , number = 3, volume = 2012, pages =

    Wolfgang F\"orstner , year = 2012, title =. Photogrammetrie -- Fernerkundung -- Geoinformatik , number = 3, volume = 2012, pages =

    2012

  7. [7]

    2024 , volume=

    Alexandre, Cyprien and Devillers, Rodolphe and Mouillot, David and Seguin, Raphael and Catry, Thibault , journal=. 2024 , volume=

    2024

  8. [8]

    2020 , publisher =

    Carion, Nicolas and Massa, Francisco and Synnaeve, Gabriel and Usunier, Nicolas and Kirillov, Alexander and Zagoruyko, Sergey , title =. 2020 , publisher =. doi:10.1007/978-3-030-58452-8\_13, Link DETR: \ ://github.com/facebookresearch/detr/tree/main?tab=readme-ov-file (last visited 07/07/2026) , booktitle =

    doi:10.1007/978-3-030-58452-8 2020

Show all 43 references
  1. [9]

    Science China Information Sciences , doi =

    Hou, Xiyue and Ao, Wei and Song, Qian and Lai, Jian and Wang, Haipeng and Xu, Feng , year =. Science China Information Sciences , doi =

  2. [10]

    2021 , volume=

    Hu, Y and Li, Y and Pan, Z , journal=. 2021 , volume=

    2021

  3. [11]

    2018 , volume=

    Huang, Lanqing and Liu, Bin and Li, Boying and Guo, Weiwei and Yu, Wenhao and Zhang, Zenghui and Yu, Wenxian , journal=. 2018 , volume=

    2018

  4. [12]

    Remote Sensing , VOLUME =

    Lei, Songlin and Lu, Dongdong and Qiu, Xiaolan and Ding, Chibiao , TITLE =. Remote Sensing , VOLUME =. 2021 , NUMBER =

    2021

  5. [13]

    Fernando Paolo and T. T. Lin and Ritwik Gupta and Bryce Goodman and Nirav Patel and Daniel Kuster and David Kroodsma and Jared Dunnmon , year=

  6. [14]

    2015 , publisher =

    Ren, Shaoqing and He, Kaiming and Girshick, Ross and Sun, Jian , title =. 2015 , publisher =

    2015

  7. [15]

    Journal of Radars. 2019. doi:10.12000/JR19097, Dataset link: \ ://radars.ac.cn/web/data/getData?newsColumnId=d25c94d7-8fe8-415f-a897-cb88657a8141&pageType=en (last visited 07/06/2026)

    2019 doi

  8. [16]

    2019 , pages=

    Zhi Tian and Chunhua Shen and Hao Chen and Tong He , journal=. 2019 , pages=

    2019

  9. [17]

    Remote Sensing , VOLUME =

    Wang, Yuanyuan and Wang, Chao and Zhang, Hong and Dong, Yingbo and Wei, Sisi , TITLE =. Remote Sensing , VOLUME =. 2019 , NUMBER =

    2019

  10. [18]

    2020 , volume=

    Wei, Shunjun and Zeng, Xiangfeng and Qu, Qizhe and Wang, Mou and Su, Hao and Shi, Jun , journal=. 2020 , volume=

    2020

  11. [19]

    Remote Sensing , VOLUME =

    Zhang, Tianwen and Zhang, Xiaoling and Li, Jianwei and Xu, Xiaowo and Wang, Baoyou and Zhan, Xu and Xu, Yanqin and Ke, Xiao and Zeng, Tianjiao and Su, Hao and Ahmad, Israr and Pan, Dece and Liu, Chang and Zhou, Yue and Shi, Jun and Wei, Shunjun , TITLE =. Remote Sensing , VOLU...

    2021

  12. [20]

    2026 , howpublished =

    ESA , title =. 2026 , howpublished =

    2026

  13. [21]

    2026 , howpublished =

    Julien Diener , title =. 2026 , howpublished =

    2026

  14. [22]

    Prasad, M. S. and Verma, Shivani and Shichkina, Yulia A. , booktitle=. 2023 , volume=

    2023

  15. [23]

    2017 , volume=

    Li, Boying and Liu, Bin and Huang, Lanqing and Guo, Weiwei and Zhang, Zenghui and Yu, Wenxian , booktitle=. 2017 , volume=

    2017

  16. [24]

    2017 , howpublished=

    Adam: A Method for Stochastic Optimization , author=. 2017 , howpublished=

    2017

  17. [25]

    Wrobel , title =

    Wolfgang F\"orstner and Bernhard P. Wrobel , title =

  18. [26]

    doi:10.1137/1.9781421407944 , address =

    Matrix Computations , author =. doi:10.1137/1.9781421407944 , address =

    doi

  19. [27]

    2011 , doi =

    Rotations in three, four, and five dimensions , author=. 2011 , doi =

    2011

  20. [28]

    Heard , year = 2006, title =

    William B. Heard , year = 2006, title =

    2006

  21. [29]

    doi:10.1002/9780470316979 , series =

    Directional Statistics , author =. doi:10.1002/9780470316979 , series =

    doi

  22. [30]

    Manual of Photogrammetry , publisher =

  23. [31]

    Method for Pointing Spacecraft While Minimizing Change in Orientation , doi =

    Patrick McKeen and Kerri Cahoy , journal =. Method for Pointing Spacecraft While Minimizing Change in Orientation , doi =

  24. [32]

    ISPRS Journal of Photogrammetry and Remote Sensing , volume = 64, number = 2, pages =

    Jochen Meidow and Christian Beder and Wolfgang F. ISPRS Journal of Photogrammetry and Remote Sensing , volume = 64, number = 2, pages =. 2009 , title =

    2009

  25. [33]

    Photogrammetric Engineering & Remote Sensing , volume = 84, number = 6, month = 6, year = 2018, pages =

    Geometric Reasoning with Uncertain Polygonal Faces , author =. Photogrammetric Engineering & Remote Sensing , volume = 84, number = 6, month = 6, year = 2018, pages =

    2018

  26. [34]

    ISPRS Journal of Photogrammetry and Remote Sensing , pages =

    Jochen Meidow , title =. ISPRS Journal of Photogrammetry and Remote Sensing , pages =. doi:10.1016/j.isprsjprs.2023.05.001 , publisher =

    2023 doi

  27. [35]

    SIAM Review , volume =

    Moler, Cleve and Van Loan, Charles , title =. SIAM Review , volume =. 2003 , doi =

    2003

  28. [36]

    Neyman and E

    J. Neyman and E. S. Pearson , year = 1933, title =. Philosophical Transactions of The Royal Society , volume = 231, pages =

    1933

  29. [37]

    Norris , journal =

    Andrew N. Norris , journal =

  30. [38]

    Journal de Mathématiques Pures et Appliquées , volume = 5, year = 1840, pages =

    Olinde Rodrigues , title =. Journal de Mathématiques Pures et Appliquées , volume = 5, year = 1840, pages =

  31. [39]

    International Journal of Chemistry, Mathematics and Physics (IJCMP) , volume = 2, number = 2, year = 2018, pages =

    One Modification which Increases Performance of N-Dimensional Rotation Matrix Generation Algorithm , author =. International Journal of Chemistry, Mathematics and Physics (IJCMP) , volume = 2, number = 2, year = 2018, pages =

    2018

  32. [40]

    B. F. Roukema , title =. Astronomy & Astrophysics , volume = 439, year = 2005, pages =

    2005

  33. [41]

    doi:10.5923/j.ajcam.20170702.04 , title =

    American Journal of Computational and Applied Mathematics , year = 2017, volume = 7, number = 2, pages =. doi:10.5923/j.ajcam.20170702.04 , title =

    2017 doi

  34. [42]

    The Journal of the Astronautical Sciences , volume = 49, number = 3, year = 2001, pages =

    On the Rigid Rotation Concept in n-Dimensional Spaces , author =. The Journal of the Astronautical Sciences , volume = 49, number = 3, year = 2001, pages =

    2001

  35. [43]

    2019 IEEE/CVF Conference on Computer Vision and Pattern Recognition (CVPR) , title =

    Yi Zhou and Connelly Barnes and Jingwan Lu and Jimei Yang and Hao Li , pages =. 2019 IEEE/CVF Conference on Computer Vision and Pattern Recognition (CVPR) , title =

    2019

Pith tools

Reviewed August 4, 2026 · model on record in the stance chip above.