Pith. sign in

REVIEW 4 major objections 6 minor 60 references

Cross-modal Ship Re-Identification via Optical and SAR Imagery: A Novel Dataset and Method

T0 review · 4 major / 6 minor · reviewed 2026-08-06 · deepseek-v4-flash

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

Pith's one-line read This paper builds the first optical-SAR ship re-identification dataset and a Vision Transformer baseline that matches ships across the two modalities at 65.9% rank-1, making all-weather, non-cooperative ship tracking from low-Earth-orbit…

desk verdict First optical-SAR ship ReID dataset is worth the field's attention, but location-confounded evaluation means the headline mAP is not yet proof of cross-modal identity matching. read the letter →

arxiv 2506.22027 v3 pith:NYPKUUTL submitted 2025-06-27 cs.CV

Han Wang , Shengyang Li , Jian Yang , Yuxuan Liu , Yixuan Lv , Zhuang Zhou This is my paper
classification cs.CV
keywords cross-modalshipre-identificationopticalandSARimageryHOSSReIDdatasetVisionTransformercontrastivepretrainingsatellitetrackingremotesensing
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

This paper's goal is to make long-term, all-weather ship tracking from low-Earth-orbit satellites practical by turning it into a cross-modal re-identification problem: given an optical image of a ship and a SAR image taken later by a different satellite, decide whether they are the same vessel. To test this, the authors built HOSS ReID, the first optical-SAR ship re-identification dataset, by tasking the Jilin-1 optical and TY-MINISAR SAR constellations to image ships in canals and ports and manually linking identities across modalities. They also propose TransOSS, a Vision Transformer that gives optical and SAR images separate tokenizers, adds learnable modality and ship-size embeddings, and is pre-trained with contrastive learning on large optical-SAR pairs. On their benchmark it reaches 57.4% mAP and 65.9% rank-1, outperforming existing cross-modal re-identification baselines. If the dataset and method hold up, ship tracking no longer depends on clear skies, daylight, or cooperative AIS broadcasts.

What carries the argument

The central object is TransOSS, a Vision Transformer with a cross-modal dual-head tokenizer: separate linear patch-embedding heads for optical and SAR inputs feed a single modality-shared transformer encoder, so the two modalities are mapped into one feature space where Euclidean distance measures similarity. Two auxiliary embeddings carry information the patch stream would otherwise lose: a learnable modality information embedding tells the encoder which sensor captured the image, and a ship-size embedding maps the estimated width, height, and aspect ratio (bounding-box size times ground sample distance) through a linear layer into the token sequence. The training recipe is the other load-bearing piece: contrastive pretraining on roughly 56K optical-SAR image pairs from SEN1-2 and DFC23 aligns paired images with a symmetric cross-entropy loss, and fine-tuning on HOSS ReID then uses ID and triplet losses. The HOSS ReID dataset itself, with 43 frames and manually linked identities, is the test bed that makes the evaluation possible.

What would settle it

Take a subset of HOSS ReID ships with nearly equal estimated length and width, run TransOSS with and without the ship-size embedding, and compare identity accuracy against labels verified with AIS trajectories for the same port and time window. If accuracy collapses when size is removed or when matched ships are size-matched, the decisive signal is bounding-box scale, not modality-invariant appearance; if labels disagree with AIS, the manual identity association underpinning every metric is unreliable.

Watch

Extended reading notes

Core claim

On its own terms, the paper establishes that cross-modal ship re-identification can be posed as a learned matching problem between optical and SAR satellite images, and that a transformer-based model can solve it better than existing visible-infrared re-identification methods. The HOSS ReID dataset is assembled from programmed imaging of the Jilin-1 optical constellation and the TY-MINISAR SAR constellation at canals and ports, with 13 image sequences, 1065 optical and 767 SAR images plus 163 distractor objects, and manually linked ship identities across modalities. On this benchmark, TransOSS reaches 57.4% mean average precision and 65.9% rank-1 accuracy overall, outperforming all compared baselines; the harder cross-modal directions still trail, at 48.9% mAP for optical-to-SAR and 38.7% mAP for SAR-to-optical. The paper reads these numbers as evidence that low-Earth-orbit optical-SAR constellations can support continuous, all-weather tracking of ships, including non-cooperative targets that do not broadcast AIS.

Load-bearing premise

The load-bearing assumption is that the manual identity links between optical and SAR images described in Section 3.2 are correct, and that the ship-size signal estimated in Section 4.2 from un-orthorectified bounding boxes reflects true vessel dimensions rather than an easy shortcut for matching ships of similar size.

Editorial extensions

If this is right

  • Cross-modal ship re-identification can now be measured on a public benchmark instead of bespoke internal data.
  • All-weather tracking no longer requires every satellite frame to come from one sensor type; optical and SAR passes can be interleaved.
  • Non-cooperative vessels without AIS become trackable from imagery alone, which existing AIS-dependent geostationary methods cannot do.
  • Larger optical-SAR paired pretraining sets are a direct lever: contrastive pretraining alone raised TransOSS mAP from 49.4% to 57.4%.
  • The cross-modal subsets remain hard at 48.9% and 38.7% mAP, so the paper's own results mark where the next improvements must land.

Reading between the lines

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

  • The paper leaves implicit that the ship-size embedding may be trading on metric scale: ships of very different lengths are easier to tell apart by size than by appearance, so the mAP gain attributed to size may not transfer to fleets of similar-sized vessels.
  • Because the dataset was captured at fixed choke points such as canals and ports, an obvious extension is to use AIS transponder logs at the same sites to verify identity labels and to scale up training pairs without further satellite tasking.
  • The recipe is not ship-specific: a dual-head tokenizer plus modality embeddings should apply to any optical-SAR association problem, such as matching vehicles, aircraft, or containers, wherever non-cooperative tracking is needed.
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

4 major / 6 minor

Summary. This manuscript introduces HOSS ReID, a new dataset for cross-modal ship re-identification using optical imagery from the Jilin-1 constellation and SAR imagery from TY-MINISAR, along with a baseline method called TransOSS. TransOSS is a Vision Transformer with a dual-head tokenizer, modality information embeddings, ship-size embeddings, and a contrastive pretraining stage on SEN1-2 and DFC23 optical-SAR pairs. The authors report that TransOSS achieves 57.4% mAP and 65.9% rank-1 on the new benchmark, outperforming several existing RGB-infrared ReID methods that were retrained on the dataset, and they position the work as the first optical-SAR ship ReID dataset for LEO-constellation ship tracking.

Significance. If the dataset labels are reliable, HOSS ReID fills a genuine gap: no public benchmark currently exists for associating ships across optical and SAR satellite imagery, and the authors state that the dataset and code are publicly available. The method is clearly described, and the ablation study in Table 3 shows that each proposed component (dual-head tokenizer, pretraining, modality embeddings, and size embeddings) contributes, with pretraining giving the largest gain. The main limitations are that the dataset cannot currently deconfound ship identity from geographic location, the identity labels lack independent verification, and the reported metrics come from single runs on a small test set. These issues directly affect the strength of the state-of-the-art claim and the benchmark's validity for the stated LEO-tracking application.

major comments (4)
  1. [Section 3.1 and Section 5.2] The dataset construction introduces a location-identity confound. All imagery is collected at prioritized ports and canals, only anchored ships are used, and each identity appears within a single one of the 13 sequences; therefore every image of a given identity shares the same scene background. The assertion in Section 3.1 that this 'makes no difference for the ship ReID task' is unsupported: a model could achieve the reported mAP by matching port/canal context or nearby vessels rather than modality-invariant ship appearance. Because the stated application is LEO tracking of moving ships across different locations, this confound directly threatens the benchmark's validity. Please provide a deconfounding analysis, such as leave-one-sequence-out evaluation, cross-sequence matching, or a background-substitution experiment, and report how performance changes; if cross-location data are unavailable, the claims and the framing as a ship-tracking benchmark should be restricted accordingly.
  2. [Section 3.2] The identity labels are created by manual association of targets across optical and SAR frames, with no AIS or independent ground-truth verification and no inter-annotator statistics. Since manual association can itself exploit location and timing cues, this leaves open the possibility that the labels encode scene identity rather than ship identity. Please report label-quality metrics (e.g., AIS cross-check on a subset, agreement rates, or a detailed description of the association protocol) or explicitly document this as a limitation.
  3. [Section 5.2 / Table 2] The reported metrics are from a single run on a test set with only 88 query tracks per modality, and no standard deviations or significance tests are given. A difference of 8.1 mAP between TransOSS and the best baseline (VersReID) could be affected by run-to-run variance, so the claim that TransOSS 'far surpasses' all other methods is not supported without repeated runs. Please report mean and standard deviation over multiple seeds (at least 3) and, where feasible, per-sequence breakdowns.
  4. [Section 4.2] The ship-size embedding is computed from bounding-box dimensions multiplied by ground sample distance on images that have not been orthorectified. Because the bounding boxes do not tightly fit ship edges and imaging geometry varies, the size feature may be noisy and could serve as a shortcut that substitutes for genuine appearance-based matching. The ablation shows that the size embedding improves mAP, but it does not show that the model is robust to geometry-induced size variation; please discuss this and, if possible, test with size features derived from orthorectified or tightly annotated boxes.
minor comments (6)
  1. [Table 2] The TransOSS row contains duplicated digits in the reported numbers (e.g., '57.457.457.4' and '65.965.965.9'), which appears to be a formatting error.
  2. [References] References [2] and [1] are identical, as are [48] and [49]; each work should be cited only once.
  3. [Section 3.1] The sentence ending '...makes no difference for the ship ReID task..' has a double period and should be corrected.
  4. [Figure 4] The caption says '(a) Distribution of ship slice heights' while the text refers to the 'length distribution of images'; the terminology should be aligned.
  5. [Section 5.4] The phrase 'operating on embeddings' is unclear; it likely refers to 'modality information embeddings' and should be made explicit.
  6. [Table 1] Please state explicitly whether the query and gallery identities are disjoint from the training identities, following the Market-1501 convention, since this is not currently stated.

Circularity Check

0 steps flagged · score 0.0 of 10

No significant circularity: the benchmark is new, the pretraining uses external datasets, and no reported result reduces by construction to a fitted value or self-cited claim.

full rationale

The paper's central results are an empirical benchmark (the new HOSS ReID dataset) and an evaluated method (TransOSS). The mAP and rank-1 numbers in Table 2 are produced by training on a train split and evaluating on held-out query and gallery splits; they are not derived from the dataset statistics or from the method's own definitions. Pretraining uses external optical-SAR pairs from SEN1-2 and DFC23, and fine-tuning uses ID and triplet losses on HOSS ReID, so the state-of-the-art claim is not statistically forced by a fitted parameter renamed as a prediction. The ship size embedding does use bounding-box-derived dimensions as an input feature, but that is a feature fed into the model, not a predicted quantity that is then reported as the ReID outcome; whether it acts as a shortcut is a benchmark-validity concern, not circularity. The manual identity association in Section 3.2 is ground-truth construction, and any label noise or location confound (each identity is tied to one of 13 port/canal sequences) is a data-quality and generalization risk, but it does not make the derivation equivalent to its inputs. The paper's unsupported assertion in Section 3.1 that anchoring 'makes no difference for the ship ReID task' should be flagged as a limitation, yet it is not a circular step. The self-citations in the related-work section (e.g., references 16, 17, 37, 56) are background references to the authors' satellite-video and SAR work and are not load-bearing for the dataset construction or the TransOSS architecture; no uniqueness theorem or unverified ansatz is imported from prior author work. No equation or claim in the paper reduces to a prior fitted value or to a self-citation chain, so the appropriate finding is no significant circularity.

Assumptions & free parameters 4 free parameters · 5 assumptions · 0 invented entities

The central claim rests mainly on the correctness of the new dataset labels and on the assumption that rough size priors and modality-shared features transfer across optical and SAR sensors. Pretraining transfer is borrowed from external datasets. No invented physical entities are introduced.

free parameters (4)
  • Modality embedding weight lambda = not reported
    Equation 2 adds lambda times the modality embedding to the token sequence; the value is chosen by hand and not reported.
  • Contrastive temperature tau = learnable, final value not reported
    Equation 4 uses a learnable tau in the symmetric cross-entropy loss, but the final value is not stated.
  • Triplet loss margin = not reported
    Equation 5 uses a pre-defined margin, but the implementation details in Section 5.1 do not give its value.
  • Input image size = 128 by 256
    Section 5.1 states that 128 by 256 'performs best in the experiments', so it is selected by validation performance.
assumptions (5)
  • domain assumption Manual identity association across optical and SAR images in HOSS ReID is correct.
    Section 3.2 states targets are manually identified and associated; any errors propagate into every reported metric.
  • domain assumption Ship size approximated from bounding box dimensions times GSD is a reliable discriminative cue.
    Section 4.2 computes iw and ih from image dimensions and GSD without orthorectification; box tightness is assumed.
  • domain assumption Optical and SAR images of the same ship share sufficient modality-invariant structure for ReID.
    Core premise of the paper in Section 1; the model is trained to extract such shared features.
  • ad hoc to paper Contrastive pretraining on SEN1-2 and DFC23 transfers to ship ReID.
    Section 4.3 uses these generic optical-SAR pairs for pretraining; transfer is assumed and only indirectly validated by the Table 3 improvement.
  • standard math Standard ViT, contrastive, ID, and triplet loss machinery behaves as expected.
    No new mathematical claims are made; standard components are adopted from cited works.

how reviews work

0 comments
Cite this review

Pith. "Pith review of Cross-modal Ship Re-Identification via Optical and SAR Imagery: A Novel Dataset and Method." pith.science (2026). https://pith.science/paper/NYPKUUTL

@misc{pith2026250622027,
  author       = {Pith},
  title        = {Pith review of: Cross-modal Ship Re-Identification via Optical and SAR Imagery: A Novel Dataset and Method},
  year         = {2026},
  howpublished = {\url{https://pith.science/paper/NYPKUUTL}},
  note         = {Machine review of arXiv:2506.22027}
}
read the original abstract

Detecting and tracking ground objects using earth observation imagery remains a significant challenge in the field of remote sensing. Continuous maritime ship tracking is crucial for applications such as maritime search and rescue, law enforcement, and shipping analysis. However, most current ship tracking methods rely on geostationary satellites or video satellites. The former offer low resolution and are susceptible to weather conditions, while the latter have short filming durations and limited coverage areas, making them less suitable for the real-world requirements of ship tracking. To address these limitations, we present the Hybrid Optical and Synthetic Aperture Radar (SAR) Ship Re-Identification Dataset (HOSS ReID dataset), designed to evaluate the effectiveness of ship tracking using low-Earth orbit constellations of optical and SAR sensors. This approach ensures shorter re-imaging cycles and enables all-weather tracking. HOSS ReID dataset includes images of the same ship captured over extended periods under diverse conditions, using different satellites of different modalities at varying times and angles. Furthermore, we propose a baseline method for cross-modal ship re-identification, TransOSS, which is built on the Vision Transformer architecture. It refines the patch embedding structure to better accommodate cross-modal tasks, incorporates additional embeddings to introduce more reference information, and employs contrastive learning to pre-train on large-scale optical-SAR image pairs, ensuring the model's ability to extract modality-invariant features. Our dataset and baseline method are publicly available on https://github.com/Alioth2000/Hoss-ReID.

Figures

Figures reproduced from arXiv: 2506.22027 by the authors.

Figure 1
Figure 1. Examples from the HOSS ReID dataset. Images in the [PITH_FULL_IMAGE:figures/full_fig_p001_1.png] view at source ↗
Figure 2
Figure 2. Schematic diagram of ship tracking method based on [PITH_FULL_IMAGE:figures/full_fig_p002_2.png] view at source ↗
Figure 4
Figure 4. (a) Distribution of ship slice heights. (b) Dataset con [PITH_FULL_IMAGE:figures/full_fig_p004_4.png] view at source ↗
Figures from the paper (3 more)
Figure 5
Figure 5. Figure 5: Framework of proposed TransOSS. Its overall structure is based on ViT, but it incorporates a dual-head tokenizer, modality [PITH_FULL_IMAGE:figures/full_fig_p005_5.png]
Figure 6
Figure 6. Figure 6: TransOSS pre-training approach. This method involves [PITH_FULL_IMAGE:figures/full_fig_p006_6.png]
Figure 7
Figure 7. Figure 7: Grad-CAM visualization of attention maps. For each set [PITH_FULL_IMAGE:figures/full_fig_p007_7.png]

Discussion (0). Sign in to comment.

Reference graph

Works this paper leans on

60 extracted references · 52 canonical work pages

  1. [1]

    Beyond appear- ance: a semantic controllable self-supervised learning frame- work for human-centric visual tasks

    Weihua Chen, Xianzhe Xu, Jian Jia, Hao Luo, Yaohua Wang, Fan Wang, Rong Jin, and Xiuyu Sun. Beyond appear- ance: a semantic controllable self-supervised learning frame- work for human-centric visual tasks. In Proceedings of the IEEE/CVF conference on computer vision and pattern recognition, pages 15050–15061, 2023. 8

    work page 2023

  2. [2]

    Beyond appear- ance: a semantic controllable self-supervised learning frame- work for human-centric visual tasks

    Weihua Chen, Xianzhe Xu, Jian Jia, Hao Luo, Yaohua Wang, Fan Wang, Rong Jin, and Xiuyu Sun. Beyond appear- ance: a semantic controllable self-supervised learning frame- work for human-centric visual tasks. In Proceedings of the IEEE/CVF conference on computer vision and pattern recognition, pages 15050–15061, 2023. 7

    work page 2023

  3. [3]

    An image is worth 16x16 words: Transformers for image recognition at scale

    Alexey Dosovitskiy. An image is worth 16x16 words: Transformers for image recognition at scale. arXiv preprint arXiv:2010.11929, 2020. 2, 7

    arXiv 2010

  4. [4]

    Cm-nas: Cross-modality neural architecture search for visible-infrared person re-identification

    Chaoyou Fu, Yibo Hu, Xiang Wu, Hailin Shi, Tao Mei, and Ran He. Cm-nas: Cross-modality neural architecture search for visible-infrared person re-identification. In Proceedings of the IEEE/CVF International Conference on Computer Vi- sion, pages 11823–11832, 2021. 7

    work page 2021

  5. [5]

    Unsuper- vised pre-training for person re-identification

    Dengpan Fu, Dongdong Chen, Jianmin Bao, Hao Yang, Lu Yuan, Lei Zhang, Houqiang Li, and Dong Chen. Unsuper- vised pre-training for person re-identification. In Proceed- ings of the IEEE/CVF conference on computer vision and pattern recognition, pages 14750–14759, 2021. 8

    work page 2021

  6. [6]

    Elimi- nate deviation with deviation for data augmentation and a general multi-modal data learning method

    Yunpeng Gong, Liqing Huang, and Lifei Chen. Elimi- nate deviation with deviation for data augmentation and a general multi-modal data learning method. arXiv preprint arXiv:2101.08533, 2021. 8

    arXiv 2021

  7. [7]

    Redet: A rotation-equivariant detector for aerial object detection

    Jiaming Han, Jian Ding, Nan Xue, and Gui-Song Xia. Redet: A rotation-equivariant detector for aerial object detection. In Proceedings of the IEEE/CVF conference on computer vi- sion and pattern recognition, pages 2786–2795, 2021. 2

    work page 2021

  8. [8]

    Transreid: Transformer-based object re- identification

    Shuting He, Hao Luo, Pichao Wang, Fan Wang, Hao Li, and Wei Jiang. Transreid: Transformer-based object re- identification. In Proceedings of the IEEE/CVF international conference on computer vision , pages 15013–15022, 2021. 5, 7

    work page 2021

Show all 60 references
  1. [9]

    Cross-modality transformer for visible-infrared person re-identification

    Kongzhu Jiang, Tianzhu Zhang, Xiang Liu, Bingqiao Qian, Yongdong Zhang, and Feng Wu. Cross-modality transformer for visible-infrared person re-identification. In European Conference on Computer Vision , pages 480–496. Springer,

  2. [10]

    Multimodal data augmentation for visual-infrared person reid with corrupted data

    Arthur Josi, Mahdi Alehdaghi, Rafael MO Cruz, and Eric Granger. Multimodal data augmentation for visual-infrared person reid with corrupted data. In Proceedings of the IEEE/CVF Winter Conference on Applications of Computer Vision, pages 32–41, 2023. 8

    2023

  3. [11]

    Maritime ship tracking based on sar constellation: a prelim- inary implementation

    Haitao Lang, Lihui Niu, Jialu Li, Wenjing Li, and Li Ma. Maritime ship tracking based on sar constellation: a prelim- inary implementation. In Image and Signal Processing for Remote Sensing XXIX, pages 260–267. SPIE, 2023. 3

    2023

  4. [12]

    Infrared-visible cross-modal person re-identification with an x modality

    Diangang Li, Xing Wei, Xiaopeng Hong, and Yihong Gong. Infrared-visible cross-modal person re-identification with an x modality. In Proceedings of the AAAI conference on artifi- cial intelligence, pages 4610–4617, 2020. 2, 3

    2020

  5. [13]

    China’s high-resolution optical remote sensing satellites and their mapping applica- tions

    Deren Li, Mi Wang, and Jie Jiang. China’s high-resolution optical remote sensing satellites and their mapping applica- tions. Geo-spatial information science , 24(1):85–94, 2021. 2

    2021

  6. [14]

    Moving ship detection based on visual saliency for video satellite

    Haichao Li and Yiyun Man. Moving ship detection based on visual saliency for video satellite. In 2016 IEEE In- ternational Geoscience and Remote Sensing Symposium (IGARSS), pages 1248–1250. IEEE, 2016. 1

    2016

  7. [15]

    All in one frame- work for multimodal re-identification in the wild

    He Li, Mang Ye, Ming Zhang, and Bo Du. All in one frame- work for multimodal re-identification in the wild. In Pro- ceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition, pages 17459–17469, 2024. 3

    2024

  8. [16]

    Recent advances in intelligent processing of satellite video: Challenges, methods, and applications

    Shengyang Li, Xian Sun, Yanfeng Gu, Yixuan Lv, Manqi Zhao, Zhuang Zhou, Weilong Guo, Yuhan Sun, Han Wang, and Jian Yang. Recent advances in intelligent processing of satellite video: Challenges, methods, and applications. IEEE Journal of Selected Topics in Applied Earth Observ...

    2023

  9. [17]

    A multitask benchmark dataset for satellite video: Object detection, tracking, and segmentation

    Shengyang Li, Zhuang Zhou, Manqi Zhao, Jian Yang, Wei- long Guo, Yixuan Lv, Longxuan Kou, Han Wang, and Yan- feng Gu. A multitask benchmark dataset for satellite video: Object detection, tracking, and segmentation. IEEE Trans- actions on Geoscience and Remote Sensing , 61:1–21...

    2023

  10. [18]

    Object tracking in satellite videos: Correlation particle filter track- ing method with motion estimation by kalman filter

    Yangfan Li, Chunjiang Bian, and Hongzhen Chen. Object tracking in satellite videos: Correlation particle filter track- ing method with motion estimation by kalman filter. IEEE Transactions on Geoscience and Remote Sensing , 60:1–12,

  11. [19]

    Deep learning-based object track- ing in satellite videos: A comprehensive survey with a new dataset

    Yuxuan Li, Licheng Jiao, Zhongjian Huang, Xin Zhang, Ruohan Zhang, Xue Song, Chenxi Tian, Zixiao Zhang, Fang Liu, Shuyuan Yang, et al. Deep learning-based object track- ing in satellite videos: A comprehensive survey with a new dataset. IEEE Geoscience and Remote Sensing Magaz...

    2022

  12. [20]

    Muls-net: A multilevel supervised network for ship tracking from low- resolution remote-sensing image sequences

    Yuan Li, Qizhi Xu, Ziyang Kong, and Wei Li. Muls-net: A multilevel supervised network for ship tracking from low- resolution remote-sensing image sequences. IEEE Transac- tions on Geoscience and Remote Sensing, 61:1–14, 2023. 1, 3

    2023

  13. [21]

    Cross- modality transformer with modality mining for visible- infrared person re-identification

    Tengfei Liang, Yi Jin, Wu Liu, and Yidong Li. Cross- modality transformer with modality mining for visible- infrared person re-identification. IEEE Transactions on Mul- timedia, 25:8432–8444, 2023. 2, 5

    2023

  14. [22]

    Bridging the gap: multi-level cross-modality joint alignment for visible-infrared person re-identification

    Tengfei Liang, Yi Jin, Wu Liu, Tao Wang, Songhe Feng, and Yidong Li. Bridging the gap: multi-level cross-modality joint alignment for visible-infrared person re-identification. IEEE Transactions on Circuits and Systems for Video Tech- nology, 2024. 7

    2024

  15. [23]

    Parameter sharing exploration and hetero-center triplet loss for visible- 9 thermal person re-identification

    Haijun Liu, Xiaoheng Tan, and Xichuan Zhou. Parameter sharing exploration and hetero-center triplet loss for visible- 9 thermal person re-identification. IEEE Transactions on Mul- timedia, 23:4414–4425, 2020. 7

    2020

  16. [24]

    Svsdet: A fine-grained recognition method for ship target using satellite video

    Shanwei Liu, Xi Bu, Mingming Xu, Hui Sheng, Zhe Zeng, and Muhammad Yasir. Svsdet: A fine-grained recognition method for ship target using satellite video. IEEE Journal of Selected Topics in Applied Earth Observations and Remote Sensing, 2024. 1

    2024

  17. [25]

    Gf-4 satellite and automatic identification system data fusion for ship tracking

    Yong Liu, Libo Yao, Wei Xiong, and Zhimin Zhou. Gf-4 satellite and automatic identification system data fusion for ship tracking. IEEE Geoscience and Remote Sensing Letters, 16(2):281–285, 2018. 1, 3

    2018

  18. [26]

    Self-supervised pre-training for transformer-based person re-identification

    Hao Luo, Pichao Wang, Yi Xu, Feng Ding, Yanxin Zhou, Fan Wang, Hao Li, and Rong Jin. Self-supervised pre-training for transformer-based person re-identification. arXiv preprint arXiv:2111.12084, 2021. 8

    2021 arXiv

  19. [27]

    Cloud removal in sentinel-2 imagery using a deep residual neural network and sar-optical data fusion

    Andrea Meraner, Patrick Ebel, Xiao Xiang Zhu, and Michael Schmitt. Cloud removal in sentinel-2 imagery using a deep residual neural network and sar-optical data fusion. ISPRS Journal of Photogrammetry and Remote Sensing , 166:333– 346, 2020. 6

    2020

  20. [28]

    H-vit: Hybrid vision transformer for multi- modal vehicle re-identification

    Wenjie Pan, Hanxiao Wu, Jianqing Zhu, Huanqiang Zeng, and Xiaobin Zhu. H-vit: Hybrid vision transformer for multi- modal vehicle re-identification. In CAAI International Con- ference on Artificial Intelligence , pages 255–267. Springer,

  21. [29]

    Learning by aligning: Visible-infrared person re- identification using cross-modal correspondences

    Hyunjong Park, Sanghoon Lee, Junghyup Lee, and Bum- sub Ham. Learning by aligning: Visible-infrared person re- identification using cross-modal correspondences. In Pro- ceedings of the IEEE/CVF international conference on com- puter vision, pages 12046–12055, 2021. 3, 5, 7

    2021

  22. [30]

    2023 ieee grss data fusion contest: Large-scale fine-grained building classification for seman- tic urban reconstruction [technical committees]

    Claudio Persello, Ronny H ¨ansch, Gemine Vivone, Kaiqiang Chen, Zhiyuan Yan, Deke Tang, Hai Huang, Michael Schmitt, and Xian Sun. 2023 ieee grss data fusion contest: Large-scale fine-grained building classification for seman- tic urban reconstruction [technical committees]. IE...

    2023

  23. [31]

    Learning transferable visual models from natural language supervi- sion

    Alec Radford, Jong Wook Kim, Chris Hallacy, Aditya Ramesh, Gabriel Goh, Sandhini Agarwal, Girish Sastry, Amanda Askell, Pamela Mishkin, Jack Clark, et al. Learning transferable visual models from natural language supervi- sion. In International conference on machine learning, ...

    2021

  24. [32]

    Implicit discriminative knowl- edge learning for visible-infrared person re-identification

    Kaijie Ren and Lei Zhang. Implicit discriminative knowl- edge learning for visible-infrared person re-identification. In Proceedings of the IEEE/CVF Conference on Computer Vi- sion and Pattern Recognition, pages 393–402, 2024. 2

    2024

  25. [33]

    The sen1-2 dataset for deep learning in sar-optical data fusion

    Michael Schmitt, Lloyd Haydn Hughes, and Xiao Xiang Zhu. The sen1-2 dataset for deep learning in sar-optical data fusion. arXiv preprint arXiv:1807.01569, 2018. 6

    2018 arXiv

  26. [34]

    Grad-cam: Visual explanations from deep networks via gradient-based localization

    Ramprasaath R Selvaraju, Michael Cogswell, Abhishek Das, Ramakrishna Vedantam, Devi Parikh, and Dhruv Batra. Grad-cam: Visual explanations from deep networks via gradient-based localization. In Proceedings of the IEEE in- ternational conference on computer vision , pages 618–626,

  27. [35]

    Improved deep metric learning with multi- class n-pair loss objective

    Kihyuk Sohn. Improved deep metric learning with multi- class n-pair loss objective. Advances in neural information processing systems, 29, 2016. 6

    2016

  28. [36]

    Training data-efficient image transformers & distillation through at- tention

    Hugo Touvron, Matthieu Cord, Matthijs Douze, Francisco Massa, Alexandre Sablayrolles, and Herv ´e J´egou. Training data-efficient image transformers & distillation through at- tention. In International conference on machine learning , pages 10347–10357. PMLR, 2021. 7

    2021

  29. [37]

    Scatter- ing information fusion network for oriented ship detection in sar images

    Han Wang, Silei Liu, Yixuan Lv, and Shengyang Li. Scatter- ing information fusion network for oriented ship detection in sar images. IEEE Geoscience and Remote Sensing Letters ,

  30. [38]

    Hrsid: A high-resolution sar images dataset for ship detection and instance segmentation

    Shunjun Wei, Xiangfeng Zeng, Qizhe Qu, Mou Wang, Hao Su, and Jun Shi. Hrsid: A high-resolution sar images dataset for ship detection and instance segmentation. Ieee Access, 8: 120234–120254, 2020. 2

    2020

  31. [39]

    Rgb-infrared cross-modality per- son re-identification

    Ancong Wu, Wei-Shi Zheng, Hong-Xing Yu, Shaogang Gong, and Jianhuang Lai. Rgb-infrared cross-modality per- son re-identification. In Proceedings of the IEEE inter- national conference on computer vision , pages 5380–5389,

  32. [40]

    Dis- cover cross-modality nuances for visible-infrared person re- identification

    Qiong Wu, Pingyang Dai, Jie Chen, Chia-Wen Lin, Yongjian Wu, Feiyue Huang, Bineng Zhong, and Rongrong Ji. Dis- cover cross-modality nuances for visible-infrared person re- identification. In Proceedings of the IEEE/CVF conference on computer vision and pattern recognition , pa...

    2021

  33. [41]

    Unsupervised visible-infrared per- son re-identification via progressive graph matching and al- ternate learning

    Zesen Wu and Mang Ye. Unsupervised visible-infrared per- son re-identification via progressive graph matching and al- ternate learning. In Proceedings of the IEEE/CVF Confer- ence on Computer Vision and Pattern Recognition , pages 9548–9558, 2023. 3

    2023

  34. [42]

    Dota: A large-scale dataset for object detection in aerial images

    Gui-Song Xia, Xiang Bai, Jian Ding, Zhen Zhu, Serge Be- longie, Jiebo Luo, Mihai Datcu, Marcello Pelillo, and Liang- pei Zhang. Dota: A large-scale dataset for object detection in aerial images. In Proceedings of the IEEE conference on computer vision and pattern recognition, ...

  35. [43]

    Yuming Xiang, Rongshu Tao, Feng Wang, Hongjian You, and Bing Han. Automatic registration of optical and sar im- ages via improved phase congruency model.IEEE Journal of Selected Topics in Applied Earth Observations and Remote Sensing, 13:5847–5861, 2020. 6

    2020

  36. [44]

    Siammdm: an adaptive fusion network with dynamic template for real-time satellite video single object tracking

    Jianwei Yang, Zongxu Pan, Ziming Wang, Bin Lei, and Yuxin Hu. Siammdm: an adaptive fusion network with dynamic template for real-time satellite video single object tracking. IEEE Transactions on Geoscience and Remote Sensing, 61:1–19, 2023. 3

    2023

  37. [45]

    Automatic ship detection in remote sensing images from google earth of complex scenes based on multiscale rotation dense feature pyramid networks

    Xue Yang, Hao Sun, Kun Fu, Jirui Yang, Xian Sun, Meng- long Yan, and Zhi Guo. Automatic ship detection in remote sensing images from google earth of complex scenes based on multiscale rotation dense feature pyramid networks. Re- mote sensing, 10(1):132, 2018. 2

    2018

  38. [46]

    R3det: Refined single-stage detector with feature refinement for ro- tating object

    Xue Yang, Junchi Yan, Ziming Feng, and Tao He. R3det: Refined single-stage detector with feature refinement for ro- tating object. In Proceedings of the AAAI conference on ar- tificial intelligence, pages 3163–3171, 2021. 2 10

    2021

  39. [47]

    A novel ship-tracking method for gf-4 satellite sequential images

    Libo Yao, Yong Liu, and You He. A novel ship-tracking method for gf-4 satellite sequential images. Sensors, 18(7): 2007, 2018. 1, 3

    2007

  40. [49]

    Deep learning for person re- identification: A survey and outlook

    Mang Ye, Jianbing Shen, Gaojie Lin, Tao Xiang, Ling Shao, and Steven CH Hoi. Deep learning for person re- identification: A survey and outlook. IEEE transactions on pattern analysis and machine intelligence, 44(6):2872–2893,

  41. [50]

    Detecting and tracking small and dense moving objects in satellite videos: A benchmark

    Qian Yin, Qingyong Hu, Hao Liu, Feng Zhang, Yingqian Wang, Zaiping Lin, Wei An, and Yulan Guo. Detecting and tracking small and dense moving objects in satellite videos: A benchmark. IEEE Transactions on Geoscience and Re- mote Sensing, 60:1–18, 2021. 1, 3

    2021

  42. [51]

    A moving ship detection and tracking method based on op- tical remote sensing images from the geostationary satellite

    Wei Yu, Hongjian You, Peng Lv, Yuxin Hu, and Bing Han. A moving ship detection and tracking method based on op- tical remote sensing images from the geostationary satellite. Sensors, 21(22):7547, 2021. 1, 3

    2021

  43. [52]

    Object knowledge distillation for joint detec- tion and tracking in satellite videos

    Wenhua Zhang, Wenjing Deng, Zhen Cui, Jia Liu, and Licheng Jiao. Object knowledge distillation for joint detec- tion and tracking in satellite videos. IEEE Transactions on Geoscience and Remote Sensing, 2024. 3

    2024

  44. [53]

    Diverse embedding expan- sion network and low-light cross-modality benchmark for visible-infrared person re-identification

    Yukang Zhang and Hanzi Wang. Diverse embedding expan- sion network and low-light cross-modality benchmark for visible-infrared person re-identification. In Proceedings of the IEEE/CVF conference on computer vision and pattern recognition, pages 2153–2162, 2023. 3, 7

    2023

  45. [54]

    Object tracking based on satellite videos: A liter- ature review

    Zhaoxiang Zhang, Chenghang Wang, Jianing Song, and Yuelei Xu. Object tracking based on satellite videos: A liter- ature review. Remote Sensing, 14(15):3674, 2022. 3

    2022

  46. [55]

    Spatial- channel enhanced transformer for visible-infrared person re- identification

    Jiaqi Zhao, Hanzheng Wang, Yong Zhou, Rui Yao, Silin Chen, and Abdulmotaleb El Saddik. Spatial- channel enhanced transformer for visible-infrared person re- identification. IEEE Transactions on Multimedia, 25:3668– 3680, 2022. 5

    2022

  47. [56]

    Mp2net: Mask propagation and mo- tion prediction network for multi-object tracking in satellite videos

    Manqi Zhao, Shengyang Li, Han Wang, Jian Yang, Yuhan Sun, and Yanfeng Gu. Mp2net: Mask propagation and mo- tion prediction network for multi-object tracking in satellite videos. IEEE Transactions on Geoscience and Remote Sens- ing, 2024. 3

    2024

  48. [57]

    Ship surveillance by integration of space-borne sar and ais–review of current research

    Zhi Zhao, Kefeng Ji, Xiangwei Xing, Huanxin Zou, and Shilin Zhou. Ship surveillance by integration of space-borne sar and ais–review of current research. The Journal of Navi- gation, 67(1):177–189, 2014. 3

    2014

  49. [58]

    Hisea-1: China’s first miniaturized commercial c-band sar satellite

    James Zheng, Qiang Chen, Xiaoyu Yan, and Weijia Ren. Hisea-1: China’s first miniaturized commercial c-band sar satellite. In IGARSS 2022-2022 IEEE International Geo- science and Remote Sensing Symposium , pages 4133–4136. IEEE, 2022. 2

    2022

  50. [59]

    Scalable person re-identification: A benchmark

    Liang Zheng, Liyue Shen, Lu Tian, Shengjin Wang, Jing- dong Wang, and Qi Tian. Scalable person re-identification: A benchmark. In Proceedings of the IEEE international con- ference on computer vision, pages 1116–1124, 2015. 4

    2015

  51. [60]

    A versa- tile framework for multi-scene person re-identification.IEEE Transactions on Pattern Analysis and Machine Intelligence, 47(3):1362–1380, 2025

    Wei-Shi Zheng, Junkai Yan, and Yi-Xing Peng. A versa- tile framework for multi-scene person re-identification.IEEE Transactions on Pattern Analysis and Machine Intelligence, 47(3):1362–1380, 2025. 7

    2025

  52. [61]

    Random erasing data augmentation

    Zhun Zhong, Liang Zheng, Guoliang Kang, Shaozi Li, and Yi Yang. Random erasing data augmentation. In Proceed- ings of the AAAI conference on artificial intelligence , pages 13001–13008, 2020. 6 11

    2020

Pith tools

Reviewed August 6, 2026 · model on record in the stance chip above.