Pith. sign in

REVIEW 5 major objections 7 minor 94 references

ViRefSAM: Visual Reference-Guided Segment Anything Model for Remote Sensing Segmentation

T0 review · 5 major / 7 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 ViRefSAM claims that visual reference images can replace manual per-image prompts for SAM, enabling automatic few-shot segmentation of unseen remote-sensing classes with consistent state-of-the-art results on three benchmarks.

desk verdict A plausible SAM-based few-shot segmenter with honest ablations but an externally confounded SOTA claim, since every ViRefSAM variant uses frozen SAM-Huge against mostly ResNet baselines. read the letter →

arxiv 2507.02294 v1 pith:ILBKJY6Y submitted 2025-07-03 cs.CV

Hanbo Bi , Yulong Xu , Ya Li , Yongqiang Mao , Boyuan Tong , Chongyang Li , Chunbo Lang , Wenhui Diao
show 3 more authors
Hongqi Wang Yingchao Feng Xian Sun
This is my paper
classification cs.CV
keywords SegmentAnythingModelremotesensingsegmentationfew-shotvisualreferencepromptingmaskattentionadapterfine-tuningmeta-learningpseudo-mask
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 tries to establish that a few annotated reference images can replace the per-image manual prompts that SAM normally requires, making remote sensing segmentation automatic for classes the model has never seen. To do this it proposes ViRefSAM, which adds two components while leaving SAM's original architecture intact: a Visual Contextual Prompt Encoder that turns reference images into object-aware prompt embeddings, and a Dynamic Target Alignment Adapter that injects class-specific reference semantics into SAM's image features. On the iSAID-5i benchmark under the 1-shot setting with a ResNet-50 backbone, ViRefSAM reports 53.06% mIoU against 51.85% for the previous best method, and it also reports gains on LoveDA-2i and COCO-20i, including over other SAM-based few-shot segmenters. If correct, this would remove the labor-intensive prompt engineering step from SAM-based remote sensing interpretation while improving generalization to novel classes.

What carries the argument

The two load-bearing components are the Visual Contextual Prompt (VCP) Encoder and the Dynamic Target Alignment (DTA) Adapter. The VCP Encoder first compresses reference features into one foreground prototype and several background prototypes obtained by a Voronoi-based split of the background, updates them via cross-attention with target features, computes a pseudo-mask by prototype similarity, and then refines learnable queries through pseudo-mask-constrained mask attention and self-attention to produce object-aware embeddings for SAM's mask decoder. The DTA Adapter, inserted in parallel in each transformer block of SAM's image encoder, down-projects target features and reference prototypes, expands the prototypes to target-image resolution using a position encoding that assigns each pixel the prototype ID from the pseudo-mask, fuses them with a 1x1 convolution, and up-projects back; only its parameters and those of the VCP encoder are trained.

What would settle it

Compute the IoU between the pseudo-mask $M^P_t$ from Eq. (3) and the ground-truth mask on iSAID-5i test folds; if that intermediate IoU is low while final mIoU stays high, the claimed reliance of mask attention and position encoding on the pseudo-mask is not the actual driver. Replacing $M^P_t$ with the oracle ground-truth mask during inference would quantify how much of the gain depends on pseudo-mask accuracy.

Watch

Extended reading notes

Core claim

The paper's central claim is that SAM's prompt-driven generalization can be redirected to unseen remote-sensing classes by conditioning it on visual references rather than handcrafted prompts. Concretely, ViRefSAM converts any reference annotation (point, scribble, box, or mask) into foreground and background prototypes, matches these against the target image to form a pseudo-mask, and uses that pseudo-mask both to constrain mask attention for object-aware prompt embeddings and to position-encode reference prototypes inside a lightweight adapter in SAM's image encoder. With only the adapter and prompt-generation modules trainable while SAM is frozen, the authors report state-of-the-art results on iSAID-5i, LoveDA-2i, and COCO-20i, and show that scribble annotations cost little accuracy, reaching 50.11% mIoU on iSAID-5i versus 53.06% with dense masks.

Load-bearing premise

The load-bearing premise is that the pseudo-mask produced by prototype matching in Eq. (3) is accurate enough to steer both the mask-attention prompt generator and the prototype position encoding; if that intermediate mask is unreliable for small, dense, or highly variable remote-sensing objects, the reported gains may come from elsewhere.

Editorial extensions

If this is right

  • Reference annotations can be points, scribbles, boxes, or masks, and scribble annotations give results close to dense masks, so the annotation burden can be lowered without a large accuracy drop.
  • One reference image suffices to beat several methods that require five reference images, which would speed up large-scale remote sensing mapping workflows.
  • The approach transfers from remote sensing to natural images, reaching 54.13% mIoU on COCO-20i and outperforming SAM-based few-shot methods such as VRP-SAM and PerSAM.
  • Freezing SAM's image encoder and tuning only the DTA adapter yields 53.06% mIoU, while full-parameter fine-tuning drops to 47.19%, suggesting parameter-efficient adaptation preserves generalization.
  • Both the visual contextual interaction and the object-aware prompt generation contribute to the gain, with removals costing 1.12% and 0.63% mIoU respectively on iSAID-5i.

Reading between the lines

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

  • The paper never directly measures the pseudo-mask's own accuracy even though both main modules consume it; an editorial next step is to report pseudo-mask IoU against ground truth and to ablate with oracle masks.
  • Table VIII suggests the DTA adapter's position encoding is a significant part of its contribution, since removing it costs 0.83 mIoU; testing the adapter with randomly assigned prototype IDs would isolate whether spatial alignment or class-semantic injection drives the gain.
  • Because ViRefSAM keeps SAM's architecture intact, it could plausibly be combined with other SAM prompt encoders or adapters, and a testable extension is to swap the VCP encoder's mask attention for a SAM-native prompt format and compare.
  • LoveDA-2i remains much harder than the object-centric datasets, with mean mIoU around 28.6%; experimenting with more background prototypes, scale augmentation, or class-balanced episodes would clarify whether the framework's gains extend to land-cover classes.
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. This paper proposes ViRefSAM, a few-shot segmentation framework built on a frozen Segment Anything Model (SAM) for remote sensing imagery. The method has two novel components: a Visual Contextual Prompt (VCP) Encoder, which computes foreground/background prototypes from a reference image, aligns them to the target image via cross-attention, derives a pseudo-mask for the target, and uses the pseudo-mask to generate object-aware prompt embeddings for SAM's mask decoder; and a Dynamic Target Alignment (DTA) Adapter, inserted in parallel into each block of SAM's image encoder, which injects reference prototypes into target features at pixel positions selected by a prototype-ID map. Training follows the episodic few-shot segmentation protocol with BCE, Dice, and a prompt-decorrelation regularization loss; inference requires only a reference image and its annotation, with no manual prompts. The paper reports experiments on iSAID-5i, LoveDA-2i, and COCO-20i, claiming consistent state-of-the-art results over both conventional few-shot segmentation (FSS) baselines and SAM-based methods such as VRP-SAM and PerSAM, together with ablations over annotation types, components, loss terms, fine-tuning strategies, and SAM encoder scales.

Significance. Reference-guided prompting of SAM for remote sensing is a timely and practically motivated direction, and the paper's internal validation is considerably more careful than is typical. In particular, the ablation in Table VI compares against a SAM+PFENet baseline that also uses the frozen SAM-Huge encoder, isolating a genuine +4.6 mIoU contribution from the two proposed components on iSAID-5i; the annotation-scheme study (Table V) and the fine-tuning-strategy study (Table VIII) are informative; and the evaluation spans two remote sensing benchmarks plus COCO-20i, supporting a generalization claim. If the concerns below are addressed, the contribution would be a useful, well-validated addition to the remote sensing few-shot segmentation literature. The paper does not ship code, and several implementation details are missing, which currently limits the reproducibility of the claimed gains.

major comments (5)
  1. [§V-B1, Tables I and X] The abstract and Section V-B1 claim that ViRefSAM 'consistently outperforms existing few-shot segmentation methods,' but the comparison in Tables I and III does not control for model scale: per Section V-A2, every ViRefSAM variant includes a frozen SAM-Huge image encoder, while nearly all listed baselines use only a ResNet-50/101 or ViT-B/16 encoder. Table X shows that the SAM encoder scale largely determines the margin: on iSAID-5i 1-shot, ViRefSAM with ViT-Base reaches 49.42 mIoU (below BAM's 50.41 and MGANet's 51.85), with ViT-Large at 51.89, and only ViT-Huge exceeds MGANet (53.06). Thus the +1.21 margin over MGANet is not attributable to the proposed framework alone. The internal ablation in Table VI does control for SAM-Huge (SAM+PFENet baseline, 48.47) and supports the effectiveness of the new components, but the external SOTA claim needs a matched-backbone comparison: either re-run the main baselines on the same frozen SAM features, or report ViRefSAM with SAM-Base/Large against the ResNet baselines across all three benchmarks.
  2. [§IV-C, Eq. (8) and Eq. (3)] Equation (8) defines the position-encoding operation Pos(·) in terms of a map \bar{M}^P_t that is never defined; the text says this quantity is 'as defined in Eq. (3),' but Eq. (3) defines M^P_t, which is stated to lie in R^{1×H×W} while the formula actually concatenates two H×W similarity maps (the background max and the foreground score). The manuscript must specify how M^P_t is converted into the per-pixel prototype-ID map \bar{M}^P_t used in Eq. (8), including the indexing convention (e.g., which ID corresponds to the foreground prototype and which to each background prototype) and how soft similarities become hard assignments. As written, the core mechanism of the DTA adapter is not reproducible.
  3. [§IV-B, Object-aware Prompt Generation] The last paragraph of Section IV-B states that the VCP encoder 'also leverages SAM's own prompt encoder to extract semantic embeddings from the pseudo-mask and its converted points and bounding boxes,' but the conversion from the pseudo-mask to these point and box prompts is never specified: no sampling rule, number of points, box extraction method, or noise strategy is given. Since these prompt embeddings are part of the input to SAM's mask decoder in the reported variants and in the K-shot experiments, the procedure must be described (or the model variant clarified) for the results to be reproducible.
  4. [§V-C, Fig. 13 and §V-A2] The hyperparameters Nq, Nb, and γ are selected using test-fold performance: the Nq sweep in Fig. 13 is evaluated on the three folds of iSAID-5i under the 1-shot protocol, and no separate validation split is mentioned anywhere in Section V. Because the test folds contain the unseen classes used to support the paper's headline claims, tuning on those folds constitutes test-set peeking and can inflate the reported mIoU. The authors should either select these hyperparameters on a validation split drawn from training classes or explicitly report the sensitivity and justify that the final choices do not rely on test-fold feedback.
  5. [§IV-B/§IV-C, pseudo-mask reliability] Both novel components depend on the pseudo-mask M^P_t and its derived prototype-ID map: it constrains the mask attention in Eq. (4) and determines the position encoding in Eq. (8). The paper never evaluates the quality of this intermediate product directly. The ablation in Table VII shows the full gain is split as +1.12 mIoU from the visual contextual interaction and +0.63 from the object-aware prompt generation, and Table VI(d) shows the DTA adapter also works when driven by PFENet's prior mask instead of the VCP pseudo-mask, so the pseudo-mask may not be the main driver of the reported gains. Please report the pseudo-mask's IoU against the ground-truth target mask (or a per-class correlation with final mIoU) to verify that the mechanism works as described in Eqs. (3)-(4) and (8).
minor comments (7)
  1. [§IV-D, Eq. (10)] In Eq. (10), the loss is written as L = BCE(Mt, Mt) + Dice(Mt, Mt) + γLreg, with the target mask Mt appearing in both arguments of each term; the first argument should be the predicted mask. This appears to be a typo but should be corrected.
  2. [Table II] In Table II, R2Net is cited as [33] in both the ResNet-50 and ResNet-101 groups; the reference list assigns [33] to DCPNet and [70] to R2Net. The citation keys should be corrected.
  3. [§I and §IV-A] The claims 'keeping SAM's original architecture intact' (Abstract) and 'without modifying SAM's core architecture' (Section IV-A) are at odds with inserting the DTA adapter into every block of the SAM image encoder as shown in Eq. (6); the phrasing should be changed to something like 'freezing SAM's original parameters with minimal architectural additions.'
  4. [Table I] The backbone column for VRP-SAM is ambiguous: VRP-SAM as published is a SAM-based method, yet it appears under ResNet-50 and ResNet-101 with different scores. Clarify whether these numbers come from a re-implementation with the stated ResNet backbone, and if so, whether the SAM-based baselines on COCO-20i (PerSAM, Matcher, VRP-SAM) were treated consistently.
  5. [§IV-B] Please state explicitly which encoder produces the features Ft and Fr used in Eqs. (1)-(4): the text calls it 'the frozen image encoder,' but Section V-A2 pairs the VCP encoder with four different backbones (ResNet-50/101, ViT-B/16, DeiT-B/16). It should also be stated which modules (VCP backbone, cross-attention projections, mask-attention queries, DTA adapter) are trainable, and how the VCP output dimension is aligned to SAM's decoder prompt dimension.
  6. [§IV-D, K-shot] The K-shot description states that K pairs of prompt embeddings are 'collectively input into SAM's mask decoder,' but it is unspecified how a variable number of reference prompt sets is combined (concatenation, averaging, or repeated decoding) and how SAM's mask decoder handles the increased token count; the 5-shot results in Table I therefore cannot be reproduced from the text.
  7. [§IV-B, Eq. (1)] The Voronoi-based background partition underlying Eq. (1) is cited to a survey and not described: please specify how the background regions are generated from the reference mask and how many regions are used, and consider releasing code, since the mask-to-prompt conversion and the partition are both implementation-critical.

Circularity Check

1 steps flagged · score 1.0 of 10

No load-bearing circularity; the only circular text is a self-referential loss equation that appears to be a typo.

  1. self definitional [Section IV-D, Eq. (10)]
    "The overall loss function of ViRefSAM is defined as: L = BCE (Mt, Mt) + Dice (Mt, Mt) + γLreg (10)"

    As printed, both BCE and Dice take the model prediction Mt as both the prediction and the target. BCE(x,x)=0 and Dice(x,x)=1, so the supervised term is the constant 1, independent of the ground-truth mask; the loss reduces to its own output and no ground-truth supervision enters. This is almost certainly a typo for the ground-truth mask, but the equation as written is self-definitional.

full rationale

ViRefSAM is an empirical architecture paper. Its pipeline (Eqs. 1-9) computes reference/background prototypes, a pseudo-mask, object-aware prompts, and adapter-modulated features, then passes them to SAM's frozen mask decoder; the final prediction is not equal by construction to any fitted input, and the pseudo-mask is an intermediate, not the claimed output. Published baseline numbers, including the authors' own DMNet and AgMTR, are external results, and the iSAID-5i/LoveDA-2i protocol from [26] has been adopted by independent groups (e.g., MGANet), so the self-citations are not load-bearing in a circular sense. The lack of public split files and the unmatched SAM-Huge backbone in the external comparison are fairness/reproducibility concerns, not circularity. The one circular artifact is Eq. (10), where the printed loss uses Mt as both prediction and ground truth; literally BCE(Mt,Mt)+Dice(Mt,Mt)=1, so the supervision term is a constant and the loss reduces to itself. This appears to be a typographical substitution of the ground-truth mask, and the reported results could not be produced by the printed loss, so it is non-load-bearing.

Assumptions & free parameters 3 free parameters · 5 assumptions · 0 invented entities

The central claim rests on a handful of domain assumptions about SAM feature transferability, the FSS protocol, and the reliability of a single reference image. The method introduces no new physical or mathematical entities. Free parameters are limited to three hyperparameters, one of which (Nq) is tuned directly on test-fold performance, which slightly inflates the reported numbers.

free parameters (3)
  • Nq (number of learnable queries) = 64
    Selected by ablation on iSAID-5i test folds (Fig. 13); mIoU peaks at Nq=64, so the value is tuned on the evaluation set.
  • gamma (regularization weight) = 0.2
    Set without an ablation; Table IX only varies the presence or absence of the regularization loss, not the weight.
  • Nb (number of background prototypes) = 5
    Chosen by hand; no sensitivity analysis is reported.
assumptions (5)
  • domain assumption SA-1B pretrained SAM features transfer to remote sensing imagery after light adapter tuning.
    The entire DTA adapter design presumes frozen SAM features are a usable base for RS objects; Section IV-C.
  • domain assumption Meta-learning on seen classes generalizes to unseen classes with disjoint labels.
    Standard FSS protocol adopted in Section III; assumes episodes sampled from training classes teach transferable prompt generation.
  • domain assumption A single reference image per class suffices to define the class semantics in diverse RS scenes.
    The method is evaluated in 1-shot and 5-shot settings; the central use case is one reference image (Section I and Fig. 4).
  • standard math Voronoi-based background partition is a valid decomposition for prototype matching.
    Used in Eq. (1) to form local background prototypes; relies on standard computational geometry.
  • domain assumption The evaluation protocol of iSAID-5i and LoveDA-2i from DMNet [26] is a fair and reproducible benchmark.
    Benchmark splits and sampling are cited to the authors' prior work [26]; no public release is referenced.

how reviews work

0 comments
Cite this review

Pith. "Pith review of ViRefSAM: Visual Reference-Guided Segment Anything Model for Remote Sensing Segmentation." pith.science (2026). https://pith.science/paper/ILBKJY6Y

@misc{pith2026250702294,
  author       = {Pith},
  title        = {Pith review of: ViRefSAM: Visual Reference-Guided Segment Anything Model for Remote Sensing Segmentation},
  year         = {2026},
  howpublished = {\url{https://pith.science/paper/ILBKJY6Y}},
  note         = {Machine review of arXiv:2507.02294}
}
abstract

The Segment Anything Model (SAM), with its prompt-driven paradigm, exhibits strong generalization in generic segmentation tasks. However, applying SAM to remote sensing (RS) images still faces two major challenges. First, manually constructing precise prompts for each image (e.g., points or boxes) is labor-intensive and inefficient, especially in RS scenarios with dense small objects or spatially fragmented distributions. Second, SAM lacks domain adaptability, as it is pre-trained primarily on natural images and struggles to capture RS-specific semantics and spatial characteristics, especially when segmenting novel or unseen classes. To address these issues, inspired by few-shot learning, we propose ViRefSAM, a novel framework that guides SAM utilizing only a few annotated reference images that contain class-specific objects. Without requiring manual prompts, ViRefSAM enables automatic segmentation of class-consistent objects across RS images. Specifically, ViRefSAM introduces two key components while keeping SAM's original architecture intact: (1) a Visual Contextual Prompt Encoder that extracts class-specific semantic clues from reference images and generates object-aware prompts via contextual interaction with target images; and (2) a Dynamic Target Alignment Adapter, integrated into SAM's image encoder, which mitigates the domain gap by injecting class-specific semantics into target image features, enabling SAM to dynamically focus on task-relevant regions. Extensive experiments on three few-shot segmentation benchmarks, including iSAID-5$^i$, LoveDA-2$^i$, and COCO-20$^i$, demonstrate that ViRefSAM enables accurate and automatic segmentation of unseen classes by leveraging only a few reference images and consistently outperforms existing few-shot segmentation methods across diverse datasets.

Figures

Figures reproduced from arXiv: 2507.02294 by the authors.

Figure 2
Figure 2. Comparison of segmentation workflow among SAM, SAM variant, and our ViRefSAM. (a) SAM: Requires manually crafted prompts (e.g., points or boxes) for each image, making it inefficient in large-scale RS segmentation. (b) SAM variant: Reuses prompts generated from a single labeled reference image per class, but lacks task-specific alignment, resulting in limited segmentation quality. (c) Our ViRefSAM: Introduces a Visu… view at source ↗
Figure 1
Figure 1. Framework comparison among SAM, SAM variant, and our ViRefSAM. (a) Original SAM: Requires manually customized prompts (e.g., points, boxes, or masks) for each individual image, which must be carefully constructed to activate the correct target. (b) SAM Variant: Attempts to reduce interaction cost by reusing prompts (e.g., a box or point) defined in one reference image across other images. However, this scheme lacks … view at source ↗
Figure 3
Figure 3. Overview of typical Few-shot Segmentation (FSS) paradigms. (a) Prototype matching-based methods [32], [49], [50], compute class prototypes from reference images and compare them with target features. (b) Feature fusion-based methods [51]–[53], aggregate reference and target features to enhance class-specific object representation. (c) Pixel matching-based methods [54]–[57], establish pixel-wise correlations between … view at source ↗
Figures from the paper (10 more)
Figure 4
Figure 4. Figure 4: Overall framework of the proposed method ViRefSAM. Without altering the original architecture of SAM, ViRefSAM introduces the reference image annotated with a specific object category and employs a Visual Contextual Prompt (VCP) Encoder to extract object-aware embeddin…
Figure 5
Figure 5. Figure 5: Detailed pipeline of the Object-aware Prompt Generator. A set of learnable queries is initialized and enhanced by the reference prototype, then refined through mask attention under pseudo-mask constraints to extract object-specific semantics from the target image. A se…
Figure 6
Figure 6. Figure 6: Detailed pipeline of the Dynamic Target Alignment (DTA) Adapter. The DTA adapter injects class-specific semantics from the reference image into the encoded features of the target image under positional constraints. This is achieved by embedding the reference prototypes…
Figure 7
Figure 7. Figure 7: Qualitative visualization of our ViRefSAM and several competing methods under the 1-shot setting on iSAID-5i benchmark. From top to bottom, each row shows: (1) reference images with ground truth masks; (2) target images with ground truth masks; (3) segmentation results…
Figure 9
Figure 9. Figure 9 [PITH_FULL_IMAGE:figures/full_fig_p012_9.png]
Figure 8
Figure 8. Figure 8: Qualitative visualization of our ViRefSAM and several competing methods under the 1-shot setting on LoveDA-2i benchmark. From left to right, each column shows: (1) reference images with ground truth masks; (2) target images with ground truth masks; (3) segmentation res…
Figure 10
Figure 10. Figure 10: Qualitative visualization of different reference annotation schemes on the iSAID-5i benchmark under the 1-shot setting. From top to bottom, reference images with ground truth annotations and four segmentation cases are provided. hinders ViRefSAM’s ability to extract m…
Figure 11
Figure 11. Figure 11: Baseline architecture of proposed ViRefSAM. It combines SAM with PFENet, where PFENet generates pseudo masks from reference images. Various prompt types (e.g., points, boxes, and masks) are randomly sampled from these predicted masks and passed through SAM’s prompt en…
Figure 13
Figure 13. Figure 13: Ablation study on the number of learnable queries (i.e., Nq) in the object-aware prompt generation from VCP encoder. The best segmentation performance is achieved at Nq = 64. ViRefSAM variants under the 1-shot setting on iSAID-5i . Overall, we observe the following: (…
Figure 14
Figure 14. Figure 14: Qualitative visualization of the Dynamic Target Alignment (DTA) Adapter. Five segmentation examples illustrate that the DTA adapter effectively guides SAM to focus on class-specific semantic regions in the current task, leading to more accurate object delineation. 3) …

Discussion (0). Sign in to comment.

Reference graph

Works this paper leans on

94 extracted references · 55 canonical work pages

  1. [1]

    Fast segmentation and classification of very high resolution remote sensing data using slic superpixels,

    O. Csillik, “Fast segmentation and classification of very high resolution remote sensing data using slic superpixels,” Remote Sensing , vol. 9, no. 3, p. 243, 2017

    2017

  2. [2]

    Remote sensing image segmentation advances: A meta-analysis,

    I. Kotaridis and M. Lazaridou, “Remote sensing image segmentation advances: A meta-analysis,” ISPRS Journal of Photogrammetry and Remote Sensing , vol. 173, pp. 309–322, 2021

    2021

  3. [3]

    Resunet-a: A deep learning framework for semantic segmentation of remotely sensed data,

    F. I. Diakogiannis, F. Waldner, P. Caccetta, and C. Wu, “Resunet-a: A deep learning framework for semantic segmentation of remotely sensed data,” ISPRS Journal of Photogrammetry and Remote Sensing , vol. 162, pp. 94–114, 2020

    2020

  4. [4]

    Deep learning semantic segmentation for land use and land cover types using landsat 8 imagery,

    W. Boonpook, Y . Tan, A. Nardkulpat, K. Torsri, P. Torteeka, P. Kamsing, U. Sawangwit, J. Pena, and M. Jainaen, “Deep learning semantic segmentation for land use and land cover types using landsat 8 imagery,” ISPRS International Journal of Geo-Information , vol. 12, no. 1, p. 14, 2023

    2023

  5. [5]

    Dynamicearthnet: Daily multi-spectral satellite dataset for semantic change segmentation,

    A. Toker, L. Kondmann, M. Weber, M. Eisenberger, A. Camero, J. Hu, A. P. Hoderlein, C ¸ . S ¸enaras, T. Davis, D. Cremers et al. , “Dynamicearthnet: Daily multi-spectral satellite dataset for semantic change segmentation,” in Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition , 2022, pp. 21 158–21 167

    2022

  6. [6]

    Remote sensing of impervious surfaces in the urban areas: Requirements, methods, and trends,

    Q. Weng, “Remote sensing of impervious surfaces in the urban areas: Requirements, methods, and trends,” Remote Sensing of Environment , vol. 117, pp. 34–49, 2012

    2012

  7. [7]

    Deep learning segmentation and classification for urban village using a worldview satellite image based on u-net,

    Z. Pan, J. Xu, Y . Guo, Y . Hu, and G. Wang, “Deep learning segmentation and classification for urban village using a worldview satellite image based on u-net,” Remote Sensing , vol. 12, no. 10, p. 1574, 2020

    2020

  8. [8]

    Msnet: A multilevel instance segmentation network for natural disaster damage assessment in aerial videos,

    X. Zhu, J. Liang, and A. Hauptmann, “Msnet: A multilevel instance segmentation network for natural disaster damage assessment in aerial videos,” in Proceedings of the IEEE/CVF winter conference on appli- cations of computer vision , 2021, pp. 2023–2032

    2021

Show all 94 references
  1. [9]

    Natural disaster damage assessment using semantic segmentation of uav imagery,

    M. H. Asad, M. M. Asim, M. N. M. Awan, and M. H. Yousaf, “Natural disaster damage assessment using semantic segmentation of uav imagery,” in 2023 International Conference on Robotics and Automation in Industry (ICRAI) . IEEE, 2023, pp. 1–7

    2023

  2. [10]

    Fully convolutional networks for semantic segmentation,

    J. Long, E. Shelhamer, and T. Darrell, “Fully convolutional networks for semantic segmentation,” in Proceedings of the IEEE conference on computer vision and pattern recognition , 2015, pp. 3431–3440

    2015

  3. [11]

    A new fully convolutional neural network for semantic segmentation of polarimetric sar imagery in complex land cover ecosys- tem,

    F. Mohammadimanesh, B. Salehi, M. Mahdianpari, E. Gill, and M. Molinier, “A new fully convolutional neural network for semantic segmentation of polarimetric sar imagery in complex land cover ecosys- tem,” ISPRS journal of photogrammetry and remote sensing , vol. 151, pp. 223–236, 2019

    2019

  4. [12]

    Semantic segmentation of remote-sensing images through fully convolutional neural networks and hierarchical probabilistic graphical models,

    M. Pastorino, G. Moser, S. B. Serpico, and J. Zerubia, “Semantic segmentation of remote-sensing images through fully convolutional neural networks and hierarchical probabilistic graphical models,” IEEE Transactions on Geoscience and Remote Sensing , vol. 60, pp. 1–16, 2022

    2022

  5. [13]

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

    A. Dosovitskiy, L. Beyer, A. Kolesnikov, D. Weissenborn, X. Zhai, T. Unterthiner, M. Dehghani, M. Minderer, G. Heigold, S. Gelly et al. , “An image is worth 16x16 words: Transformers for image recognition at scale,” arXiv preprint arXiv:2010.11929 , 2020

    2010 arXiv

  6. [14]

    Unetformer: A unet-like transformer for efficient semantic segmentation of remote sensing urban scene imagery,

    L. Wang, R. Li, C. Zhang, S. Fang, C. Duan, X. Meng, and P. M. Atkinson, “Unetformer: A unet-like transformer for efficient semantic segmentation of remote sensing urban scene imagery,” ISPRS Journal of Photogrammetry and Remote Sensing , vol. 190, pp. 196–214, 2022

    2022

  7. [15]

    Masked-attention mask transformer for universal image segmentation,

    B. Cheng, I. Misra, A. G. Schwing, A. Kirillov, and R. Girdhar, “Masked-attention mask transformer for universal image segmentation,” in Proceedings of the IEEE/CVF conference on computer vision and pattern recognition, 2022, pp. 1290–1299

    2022

  8. [16]

    Segment anything,

    A. Kirillov, E. Mintun, N. Ravi, H. Mao, C. Rolland, L. Gustafson, T. Xiao, S. Whitehead, A. C. Berg, W.-Y . Loet al., “Segment anything,” in Proceedings of the IEEE/CVF international conference on computer vision, 2023, pp. 4015–4026

    2023

  9. [17]

    Fast segment anything,

    X. Zhao, W. Ding, Y . An, Y . Du, T. Yu, M. Li, M. Tang, and J. Wang, “Fast segment anything,” arXiv preprint arXiv:2306.12156 , 2023

    2023 arXiv

  10. [18]

    Sam-adapter: Adapting segment anything in underperformed scenes,

    T. Chen, L. Zhu, C. Deng, R. Cao, Y . Wang, S. Zhang, Z. Li, L. Sun, Y . Zang, and P. Mao, “Sam-adapter: Adapting segment anything in underperformed scenes,” in Proceedings of the IEEE/CVF International Conference on Computer Vision , 2023, pp. 3367–3375

    2023

  11. [19]

    Medical sam adapter: Adapting segment anything model for medical image segmentation,

    J. Wu, Z. Wang, M. Hong, W. Ji, H. Fu, Y . Xu, M. Xu, and Y . Jin, “Medical sam adapter: Adapting segment anything model for medical image segmentation,” Medical image analysis, vol. 102, p. 103547, 2025

    2025

  12. [20]

    Samrs: Scaling-up remote sensing segmentation dataset with segment anything model,

    D. Wang, J. Zhang, B. Du, M. Xu, L. Liu, D. Tao, and L. Zhang, “Samrs: Scaling-up remote sensing segmentation dataset with segment anything model,” Advances in Neural Information Processing Systems , vol. 36, pp. 8815–8827, 2023

    2023

  13. [21]

    Rsprompter: Learning to prompt for remote sensing instance seg- mentation based on visual foundation model,

    K. Chen, C. Liu, H. Chen, H. Zhang, W. Li, Z. Zou, and Z. Shi, “Rsprompter: Learning to prompt for remote sensing instance seg- mentation based on visual foundation model,” IEEE Transactions on Geoscience and Remote Sensing , vol. 62, pp. 1–17, 2024

    2024

  14. [22]

    Ringmo-sam: A foundation model for segment anything in multimodal remote-sensing images,

    Z. Yan, J. Li, X. Li, R. Zhou, W. Zhang, Y . Feng, W. Diao, K. Fu, and X. Sun, “Ringmo-sam: A foundation model for segment anything in multimodal remote-sensing images,” IEEE Transactions on Geoscience and Remote Sensing , vol. 61, pp. 1–16, 2023

    2023

  15. [23]

    Sam- assisted remote sensing imagery semantic segmentation with object and boundary constraints,

    X. Ma, Q. Wu, X. Zhao, X. Zhang, M.-O. Pun, and B. Huang, “Sam- assisted remote sensing imagery semantic segmentation with object and boundary constraints,” IEEE Transactions on Geoscience and Remote Sensing, 2024

    2024

  16. [24]

    Sam enhanced semantic segmentation for remote sensing imagery without additional training,

    Y . Qiao, B. Zhong, B. Du, H. Cai, J. Jiang, Q. Liu, A. Yang, J. Wu, and X. Wang, “Sam enhanced semantic segmentation for remote sensing imagery without additional training,” IEEE Transactions on Geoscience and Remote Sensing , 2025

    2025

  17. [25]

    Matching net- works for one shot learning,

    O. Vinyals, C. Blundell, T. Lillicrap, D. Wierstra et al. , “Matching net- works for one shot learning,” Advances in neural information processing systems, vol. 29, 2016

    2016

  18. [26]

    Not just learning from others but relying on yourself: A new perspective on few-shot segmentation in remote sensing,

    H. Bi, Y . Feng, Z. Yan, Y . Mao, W. Diao, H. Wang, and X. Sun, “Not just learning from others but relying on yourself: A new perspective on few-shot segmentation in remote sensing,” IEEE Transactions on Geoscience and Remote Sensing , 2023

    2023

  19. [27]

    Learning to compare: Relation network for few-shot learning,

    F. Sung, Y . Yang, L. Zhang, T. Xiang, P. H. Torr, and T. M. Hospedales, “Learning to compare: Relation network for few-shot learning,” in Proceedings of the IEEE conference on computer vision and pattern recognition, 2018, pp. 1199–1208

    2018

  20. [28]

    Generalizing from a few examples: A survey on few-shot learning,

    Y . Wang, Q. Yao, J. T. Kwok, and L. M. Ni, “Generalizing from a few examples: A survey on few-shot learning,” ACM computing surveys (csur), vol. 53, no. 3, pp. 1–34, 2020. JOURNAL OF LATEX CLASS FILES, VOL. 14, NO. 8, AUGUST 2021 17

    2020

  21. [29]

    Model-agnostic meta-learning for fast adaptation of deep networks,

    C. Finn, P. Abbeel, and S. Levine, “Model-agnostic meta-learning for fast adaptation of deep networks,” in International conference on machine learning . PMLR, 2017, pp. 1126–1135

    2017

  22. [30]

    Vl-adapter: Parameter-efficient transfer learning for vision-and-language tasks,

    Y .-L. Sung, J. Cho, and M. Bansal, “Vl-adapter: Parameter-efficient transfer learning for vision-and-language tasks,” in Proceedings of the IEEE/CVF conference on computer vision and pattern recognition , 2022, pp. 5227–5237

    2022

  23. [31]

    Panet: Few-shot image semantic segmentation with prototype alignment,

    K. Wang, J. H. Liew, Y . Zou, D. Zhou, and J. Feng, “Panet: Few-shot image semantic segmentation with prototype alignment,” in proceedings of the IEEE/CVF international conference on computer vision , 2019, pp. 9197–9206

    2019

  24. [32]

    Part-aware prototype net- work for few-shot semantic segmentation,

    Y . Liu, X. Zhang, S. Zhang, and X. He, “Part-aware prototype net- work for few-shot semantic segmentation,” in Computer Vision–ECCV 2020: 16th European Conference, Glasgow, UK, August 23–28, 2020, Proceedings, Part IX 16 . Springer, 2020, pp. 142–158

    2020

  25. [33]

    Few-shot segmentation via divide-and-conquer proxies,

    C. Lang, G. Cheng, B. Tu, and J. Han, “Few-shot segmentation via divide-and-conquer proxies,” International Journal of Computer Vision , vol. 132, no. 1, pp. 261–283, 2024

    2024

  26. [34]

    Semantic image segmentation with deep convolutional nets and fully connected crfs,

    L.-C. Chen, G. Papandreou, I. Kokkinos, K. Murphy, and A. L. Yuille, “Semantic image segmentation with deep convolutional nets and fully connected crfs,” arXiv preprint arXiv:1412.7062 , 2014

    2014 arXiv

  27. [35]

    Pyramid scene parsing network,

    H. Zhao, J. Shi, X. Qi, X. Wang, and J. Jia, “Pyramid scene parsing network,” in Proceedings of the IEEE conference on computer vision and pattern recognition , 2017, pp. 2881–2890

    2017

  28. [36]

    Deeplab: Semantic image segmentation with deep convolutional nets, atrous convolution, and fully connected crfs,

    L.-C. Chen, G. Papandreou, I. Kokkinos, K. Murphy, and A. L. Yuille, “Deeplab: Semantic image segmentation with deep convolutional nets, atrous convolution, and fully connected crfs,” IEEE transactions on pattern analysis and machine intelligence , vol. 40, no. 4, pp. 834–848, 2017

    2017

  29. [37]

    Cross fusion net: A fast semantic segmentation network for small-scale semantic information capturing in aerial scenes,

    C. Peng, K. Zhang, Y . Ma, and J. Ma, “Cross fusion net: A fast semantic segmentation network for small-scale semantic information capturing in aerial scenes,” IEEE Transactions on Geoscience and Remote Sensing , vol. 60, pp. 1–13, 2021

    2021

  30. [38]

    Npaloss: Neighboring pixel affinity loss for semantic segmentation in high-resolution aerial imagery,

    Y . Feng, W. Diao, X. Sun, J. Li, K. Chen, K. Fu, and X. Gao, “Npaloss: Neighboring pixel affinity loss for semantic segmentation in high-resolution aerial imagery,” ISPRS Annals of the Photogrammetry, Remote Sensing and Spatial Information Sciences , vol. 2, pp. 475–482, 2020

    2020

  31. [39]

    Improv- ing semantic segmentation in aerial imagery via graph reasoning and disentangled learning,

    R. Niu, X. Sun, Y . Tian, W. Diao, Y . Feng, and K. Fu, “Improv- ing semantic segmentation in aerial imagery via graph reasoning and disentangled learning,” IEEE Transactions on Geoscience and Remote Sensing, vol. 60, pp. 1–18, 2022

    2022

  32. [40]

    Segformer: Simple and efficient design for semantic segmentation with transformers,

    E. Xie, W. Wang, Z. Yu, A. Anandkumar, J. M. Alvarez, and P. Luo, “Segformer: Simple and efficient design for semantic segmentation with transformers,” Advances in neural information processing systems , vol. 34, pp. 12 077–12 090, 2021

    2021

  33. [41]

    Swin transformer embedding unet for remote sensing image semantic segmen- tation,

    X. He, Y . Zhou, J. Zhao, D. Zhang, R. Yao, and Y . Xue, “Swin transformer embedding unet for remote sensing image semantic segmen- tation,” IEEE transactions on geoscience and remote sensing , vol. 60, pp. 1–15, 2022

    2022

  34. [42]

    Enhancing multiscale representations with transformer for remote sensing image semantic segmentation,

    T. Xiao, Y . Liu, Y . Huang, M. Li, and G. Yang, “Enhancing multiscale representations with transformer for remote sensing image semantic segmentation,” IEEE Transactions on Geoscience and Remote Sensing , vol. 61, pp. 1–16, 2023

    2023

  35. [43]

    Rethinking transformers for semantic segmentation of remote sensing images,

    Y . Liu, Y . Zhang, Y . Wang, and S. Mei, “Rethinking transformers for semantic segmentation of remote sensing images,” IEEE Transactions on Geoscience and Remote Sensing , vol. 61, pp. 1–15, 2023

    2023

  36. [44]

    Segment anything in medical images,

    J. Ma, Y . He, F. Li, L. Han, C. You, and B. Wang, “Segment anything in medical images,” Nature Communications, vol. 15, no. 1, p. 654, 2024

    2024

  37. [45]

    Inpaint anything: Segment anything meets image inpainting,

    T. Yu, R. Feng, R. Feng, J. Liu, X. Jin, W. Zeng, and Z. Chen, “Inpaint anything: Segment anything meets image inpainting,” arXiv preprint arXiv:2304.06790, 2023

    2023 arXiv

  38. [46]

    Edit everything: A text-guided generative system for images editing,

    D. Xie, R. Wang, J. Ma, C. Chen, H. Lu, D. Yang, F. Shi, and X. Lin, “Edit everything: A text-guided generative system for images editing,” arXiv preprint arXiv:2304.14006 , 2023

    2023 arXiv

  39. [47]

    Segment and track anything,

    Y . Cheng, L. Li, Y . Xu, X. Li, Z. Yang, W. Wang, and Y . Yang, “Segment and track anything,” arXiv preprint arXiv:2305.06558 , 2023

    2023 arXiv

  40. [48]

    Segment anything meets point tracking,

    F. Raji ˇc, L. Ke, Y .-W. Tai, C.-K. Tang, M. Danelljan, and F. Yu, “Segment anything meets point tracking,” in 2025 IEEE/CVF Winter Conference on Applications of Computer Vision (WACV) . IEEE, 2025, pp. 9302–9311

    2025

  41. [49]

    Sg-one: Similarity guidance network for one-shot semantic segmentation,

    X. Zhang, Y . Wei, Y . Yang, and T. S. Huang, “Sg-one: Similarity guidance network for one-shot semantic segmentation,” IEEE Trans. Cybern., vol. 50, no. 9, pp. 3855–3865, 2020

    2020

  42. [50]

    Self-support few-shot semantic segmentation,

    Q. Fan, W. Pei, Y .-W. Tai, and C.-K. Tang, “Self-support few-shot semantic segmentation,” in Proc. Eur . Conf. Comput. Vis. Springer, 2022, pp. 701–719

    2022

  43. [51]

    Hypercorrelation squeeze for few-shot segmentation,

    J. Min, D. Kang, and M. Cho, “Hypercorrelation squeeze for few-shot segmentation,” in Proc. IEEE Int. Conf. Comput. Vis. , 2021, pp. 6941– 6952

    2021

  44. [52]

    Holistic prototype activation for few- shot segmentation,

    G. Cheng, C. Lang, and J. Han, “Holistic prototype activation for few- shot segmentation,” IEEE Transactions on Pattern Analysis and Machine Intelligence, vol. 45, no. 4, pp. 4650–4666, 2022

    2022

  45. [53]

    Learning non- target knowledge for few-shot semantic segmentation,

    Y . Liu, N. Liu, Q. Cao, X. Yao, J. Han, and L. Shao, “Learning non- target knowledge for few-shot semantic segmentation,” inProceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition , 2022, pp. 11 573–11 582

    2022

  46. [54]

    Few-shot segmentation via cycle-consistent transformer,

    G. Zhang, G. Kang, Y . Yang, and Y . Wei, “Few-shot segmentation via cycle-consistent transformer,” Advances in Neural Information Process- ing Systems , vol. 34, pp. 21 984–21 996, 2021

    2021

  47. [55]

    Adaptive agent transformer for few-shot segmentation,

    Y . Wang, R. Sun, Z. Zhang, and T. Zhang, “Adaptive agent transformer for few-shot segmentation,” in European Conference on Computer Vision. Springer, 2022, pp. 36–52

    2022

  48. [56]

    Self-calibrated cross attention network for few-shot segmentation,

    Q. Xu, W. Zhao, G. Lin, and C. Long, “Self-calibrated cross attention network for few-shot segmentation,” in Proceedings of the IEEE/CVF International Conference on Computer Vision , 2023, pp. 655–665

    2023

  49. [57]

    Hierarchical dense correlation distillation for few-shot segmentation,

    B. Peng, Z. Tian, X. Wu, C. Wang, S. Liu, J. Su, and J. Jia, “Hierarchical dense correlation distillation for few-shot segmentation,” in Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recogni- tion, 2023, pp. 23 641–23 651

    2023

  50. [58]

    Mesam: Multiscale enhanced segment anything model for optical remote sensing images,

    X. Zhou, F. Liang, L. Chen, H. Liu, Q. Song, G. Vivone, and J. Chanus- sot, “Mesam: Multiscale enhanced segment anything model for optical remote sensing images,” IEEE Transactions on Geoscience and Remote Sensing, vol. 62, pp. 1–15, 2024

    2024

  51. [59]

    Sam-rsis: Progressively adapting sam with box prompting to remote sensing image instance segmenta- tion,

    M. Luo, T. Zhang, S. Wei, and S. Ji, “Sam-rsis: Progressively adapting sam with box prompting to remote sensing image instance segmenta- tion,” IEEE Transactions on Geoscience and Remote Sensing , 2024

    2024

  52. [60]

    Road-sam: Adapting the seg- ment anything model to road extraction from large very-high-resolution optical remote sensing images,

    W. Feng, F. Guan, C. Sun, and W. Xu, “Road-sam: Adapting the seg- ment anything model to road extraction from large very-high-resolution optical remote sensing images,” IEEE Geoscience and Remote Sensing Letters, vol. 21, pp. 1–5, 2024

    2024

  53. [61]

    Tpp-sam: A trajectory points prompting segment anything model for zero-shot road extraction from high-resolution remote sensing imagery,

    T. Wu, Y . Hu, J. Qin, X. Lin, and Y . Wan, “Tpp-sam: A trajectory points prompting segment anything model for zero-shot road extraction from high-resolution remote sensing imagery,” IEEE Journal of Selected Topics in Applied Earth Observations and Remote Sensing , 2025

    2025

  54. [62]

    Multi-view remote sensing image segmentation with sam priors,

    Z. Qi, C. Liu, Z. Liu, H. Chen, Y . Wu, Z. Zou, and Z. Shi, “Multi-view remote sensing image segmentation with sam priors,” in IGARSS 2024- 2024 IEEE International Geoscience and Remote Sensing Symposium . IEEE, 2024, pp. 8446–8449

    2024

  55. [63]

    Adapting segment anything model for change detection in vhr remote sensing images,

    L. Ding, K. Zhu, D. Peng, H. Tang, K. Yang, and L. Bruzzone, “Adapting segment anything model for change detection in vhr remote sensing images,” IEEE Transactions on Geoscience and Remote Sensing , 2024

    2024

  56. [64]

    One-shot learning for semantic segmentation,

    A. Shaban, S. Bansal, Z. Liu, I. Essa, and B. Boots, “One-shot learning for semantic segmentation,” arXiv preprint arXiv:1709.03410 , 2017

    2017 arXiv

  57. [65]

    Mianet: Aggregating unbiased instance and general information for few-shot semantic seg- mentation,

    Y . Yang, Q. Chen, Y . Feng, and T. Huang, “Mianet: Aggregating unbiased instance and general information for few-shot semantic seg- mentation,” in Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition , 2023, pp. 7131–7140

    2023

  58. [66]

    Intermediate prototype mining transformer for few-shot semantic segmentation,

    Y . Liu, N. Liu, X. Yao, and J. Han, “Intermediate prototype mining transformer for few-shot semantic segmentation,” Advances in Neural Information Processing Systems , vol. 35, pp. 38 020–38 031, 2022

    2022

  59. [67]

    Scale-aware detailed matching for few-shot aerial image semantic segmentation,

    X. Yao, Q. Cao, X. Feng, G. Cheng, and J. Han, “Scale-aware detailed matching for few-shot aerial image semantic segmentation,”IEEE Trans- actions on Geoscience and Remote Sensing , vol. 60, pp. 1–11, 2021

    2021

  60. [68]

    Dmml-net: Deep metametric learning for few-shot geographic object segmentation in remote sensing imagery,

    B. Wang, Z. Wang, X. Sun, H. Wang, and K. Fu, “Dmml-net: Deep metametric learning for few-shot geographic object segmentation in remote sensing imagery,” IEEE Transactions on Geoscience and Remote Sensing, vol. 60, pp. 1–18, 2021

    2021

  61. [69]

    Few-shot segmentation of remote sensing images using deep metric learning,

    X. Jiang, N. Zhou, and X. Li, “Few-shot segmentation of remote sensing images using deep metric learning,” IEEE Geoscience and Remote Sensing Letters , vol. 19, pp. 1–5, 2022

    2022

  62. [70]

    Global rectification and decoupled registration for few-shot segmentation in remote sensing imagery,

    C. Lang, G. Cheng, B. Tu, and J. Han, “Global rectification and decoupled registration for few-shot segmentation in remote sensing imagery,” IEEE Transactions on Geoscience and Remote Sensing , 2023

    2023

  63. [71]

    Progressive parsing and commonality distillation for few-shot remote sensing segmentation,

    C. Lang, J. Wang, G. Cheng, B. Tu, and J. Han, “Progressive parsing and commonality distillation for few-shot remote sensing segmentation,” IEEE Transactions on Geoscience and Remote Sensing , vol. 61, pp. 1– 10, 2023

    2023

  64. [72]

    Multi-granularity aggregation network for remote sensing few-shot segmentation,

    S.-F. Peng, G.-S. Xie, F. Zhao, X. Shu, and Q. Liu, “Multi-granularity aggregation network for remote sensing few-shot segmentation,” IEEE Transactions on Geoscience and Remote Sensing , pp. 1–1, 2024

    2024

  65. [73]

    V oronoi diagrams—a survey of a fundamental geo- metric data structure,

    F. Aurenhammer, “V oronoi diagrams—a survey of a fundamental geo- metric data structure,” ACM computing surveys (CSUR) , vol. 23, no. 3, pp. 345–405, 1991. JOURNAL OF LATEX CLASS FILES, VOL. 14, NO. 8, AUGUST 2021 18

    1991

  66. [74]

    Adaptive prototype learning and allocation for few-shot segmentation,

    G. Li, V . Jampani, L. Sevilla-Lara, D. Sun, J. Kim, and J. Kim, “Adaptive prototype learning and allocation for few-shot segmentation,” in Proceedings of the IEEE/CVF conference on computer vision and pattern recognition, 2021, pp. 8334–8343

    2021

  67. [75]

    Prior guided feature enrichment network for few-shot segmentation,

    Z. Tian, H. Zhao, M. Shu, Z. Yang, R. Li, and J. Jia, “Prior guided feature enrichment network for few-shot segmentation,” IEEE Transactions on Pattern Analysis & Machine Intelligence, vol. 44, no. 02, pp. 1050–1065, 2022

    2022

  68. [76]

    Base and meta: A new perspective on few-shot segmentation,

    C. Lang, G. Cheng, B. Tu, C. Li, and J. Han, “Base and meta: A new perspective on few-shot segmentation,” IEEE Transactions on Pattern Analysis and Machine Intelligence , 2023

    2023

  69. [77]

    Vrp-sam: Sam with visual reference prompt,

    Y . Sun, J. Chen, S. Zhang, X. Zhang, Q. Chen, G. Zhang, E. Ding, J. Wang, and Z. Li, “Vrp-sam: Sam with visual reference prompt,” in Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition, 2024, pp. 23 565–23 574

    2024

  70. [78]

    Feature-proxy transformer for few-shot segmentation,

    J.-W. Zhang, Y . Sun, Y . Yang, and W. Chen, “Feature-proxy transformer for few-shot segmentation,” Advances in Neural Information Processing Systems, vol. 35, pp. 6575–6588, 2022

    2022

  71. [79]

    Agmtr: Agent mining transformer for few-shot segmentation in remote sensing,

    H. Bi, Y . Feng, Y . Mao, J. Pei, W. Diao, H. Wang, and X. Sun, “Agmtr: Agent mining transformer for few-shot segmentation in remote sensing,” International Journal of Computer Vision , pp. 1–28, 2024

    2024

  72. [80]

    isaid: A large-scale dataset for instance segmentation in aerial images,

    S. Waqas Zamir, A. Arora, A. Gupta, S. Khan, G. Sun, F. Shahbaz Khan, F. Zhu, L. Shao, G.-S. Xia, and X. Bai, “isaid: A large-scale dataset for instance segmentation in aerial images,” inProceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition Workshop...

    2019

  73. [81]

    LoveDA: A remote sensing land-cover dataset for domain adaptive semantic segmentation,

    J. Wang, Z. Zheng, A. Ma, X. Lu, and Y . Zhong, “LoveDA: A remote sensing land-cover dataset for domain adaptive semantic segmentation,” in Proceedings of the Neural Information Processing Systems Track on Datasets and Benchmarks , J. Vanschoren and S. Yeung, Eds., vol. 1, 202...

    2021

  74. [82]

    Microsoft coco: Common objects in context,

    T.-Y . Lin, M. Maire, S. Belongie, J. Hays, P. Perona, D. Ramanan, P. Doll ´ar, and C. L. Zitnick, “Microsoft coco: Common objects in context,” in Computer Vision–ECCV 2014: 13th European Conference, Zurich, Switzerland, September 6-12, 2014, Proceedings, Part V 13 . Springer,...

    2014

  75. [83]

    Self-guided and cross-guided learning for few-shot segmentation,

    B. Zhang, J. Xiao, and T. Qin, “Self-guided and cross-guided learning for few-shot segmentation,” in Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition , 2021, pp. 8312–8321

    2021

  76. [84]

    Pytorch: An imperative style, high-performance deep learning library,

    A. Paszke, S. Gross, F. Massa, A. Lerer, J. Bradbury, G. Chanan, T. Killeen, Z. Lin, N. Gimelshein, L. Antiga et al. , “Pytorch: An imperative style, high-performance deep learning library,” Advances in neural information processing systems , vol. 32, 2019

    2019

  77. [85]

    Training data-efficient image transformers & distillation through attention,

    H. Touvron, M. Cord, M. Douze, F. Massa, A. Sablayrolles, and H. J ´egou, “Training data-efficient image transformers & distillation through attention,” in International conference on machine learning . PMLR, 2021, pp. 10 347–10 357

    2021

  78. [86]

    Imagenet: A large-scale hierarchical image database,

    J. Deng, W. Dong, R. Socher, L.-J. Li, K. Li, and L. Fei-Fei, “Imagenet: A large-scale hierarchical image database,” in 2009 IEEE conference on computer vision and pattern recognition . Ieee, 2009, pp. 248–255

    2009

  79. [87]

    Seggpt: Towards segmenting everything in context,

    X. Wang, X. Zhang, Y . Cao, W. Wang, C. Shen, and T. Huang, “Seggpt: Towards segmenting everything in context,” in Proceedings of the IEEE/CVF International Conference on Computer Vision , 2023, pp. 1130–1140

    2023

  80. [88]

    Personalize segment anything model with one shot,

    R. Zhang, Z. Jiang, Z. Guo, S. Yan, J. Pan, X. Ma, H. Dong, P. Gao, and H. Li, “Personalize segment anything model with one shot,” arXiv preprint arXiv:2305.03048, 2023

    2023 arXiv

  81. [89]

    Llafs: When large language models meet few-shot segmentation,

    L. Zhu, T. Chen, D. Ji, J. Ye, and J. Liu, “Llafs: When large language models meet few-shot segmentation,” in Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition , 2024, pp. 3065–3075

    2024

  82. [90]

    Images speak in images: A generalist painter for in-context visual learning,

    X. Wang, W. Wang, Y . Cao, C. Shen, and T. Huang, “Images speak in images: A generalist painter for in-context visual learning,” in Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition, 2023, pp. 6830–6839

    2023

  83. [91]

    Dual branch multi-level semantic learning for few-shot segmentation,

    Y . Chen, R. Jiang, Y . Zheng, B. Sheng, Z.-X. Yang, and E. Wu, “Dual branch multi-level semantic learning for few-shot segmentation,” IEEE Transactions on Image Processing , vol. 33, pp. 1432–1447, 2024

    2024

  84. [92]

    Hybrid mamba for few-shot segmentation,

    Q. Xu, X. Liu, L. Zhu, G. Lin, C. Long, Z. Li, and R. Zhao, “Hybrid mamba for few-shot segmentation,” Advances in Neural Information Processing Systems, vol. 37, pp. 73 858–73 883, 2024

    2024

  85. [93]

    Matching anything by segmenting anything,

    S. Li, L. Ke, M. Danelljan, L. Piccinelli, M. Segu, L. Van Gool, and F. Yu, “Matching anything by segmenting anything,” in Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition , 2024, pp. 18 963–18 973

    2024

  86. [94]

    Prompt-and-transfer: Dynamic class-aware enhancement for few-shot segmentation,

    H. Bi, Y . Feng, W. Diao, P. Wang, Y . Mao, K. Fu, H. Wang, and X. Sun, “Prompt-and-transfer: Dynamic class-aware enhancement for few-shot segmentation,” IEEE Transactions on Pattern Analysis and Machine Intelligence, 2024

    2024

Pith tools

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