Pith. sign in

REVIEW 3 major objections 5 minor 37 references

Checkmate: interpretable and explainable RSVQA is the endgame

T0 review · 3 major / 5 minor · reviewed 2026-08-15 · deepseek-v4-flash

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

Pith's one-line read Checkmate builds the evidence cells into RSVQA answers, reaching 50.9% average accuracy while naming the cells behind each answer.

desk verdict Chessboard is a solid dataset contribution, but Checkmate's explainability claim is unsupported: the predicted cells are a parallel output, not the cause of the answer, and the Blind model already solves much of the cell task from text alone. read the letter →

arxiv 2508.13086 v1 pith:MEZHXZTX submitted 2025-08-18 cs.CV

Lucrezia Tosato , Christel Tartini Chappuis , Syrielle Montariol , Flora Weissgerber , Sylvain Lobry , Devis Tuia This is my paper · ORCID
classification cs.CV
keywords RemotesensingvisualquestionansweringinterpretabilityexplainabilitysemanticsegmentationcellgroundingbiasmitigationChessboarddataset
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 argues that remote-sensing visual question answering, which means asking a model questions about satellite images and getting text answers, can be made transparent from the inside instead of being explained after the fact. Its proposal, Checkmate, first maps the image into a semantic segmentation, then converts that map into a short text summary for each of 16 image cells, and a language model reads the question and the summary together to predict both the answer and the cells on which the answer rests. To train and test this, the authors built Chessboard, a dataset of 3,123,253 balanced questions over reBEN satellite patches, with every answer annotated with the image cells that justify it. Checkmate with a SegFormer segmenter reaches an average answer accuracy of 50.9%, while predicting the supporting cells with an 82.5% micro-F1 score. If these results hold, users of RSVQA systems would be able to see not only what the model says but where in the image it looked to say it.

What carries the argument

The mechanism is the cell-level textual summary: the image is divided into a 4-by-4 grid labeled a1 to d4, and for each cell the segmentation map lists the land-cover classes present in that cell, discarding classes with fewer than 30 pixels; total class pixel counts are binned into 290 area ranges, and the resulting structured sentences are concatenated with the question and fed to DistilBERT, whose two heads predict the answer among 335 classes and the 16 relevant cells as a multi-label classification. This single representation does double duty: it is the only visual information the language model sees, and it is also the explanation the model outputs, since the predicted cells name where the model looked.

What would settle it

Construct a Chessboard test where a class covers 25 pixels in one cell, below the 30-pixel threshold, but is present elsewhere, and ask the model a question that forces attention to that cell; if the predicted cells consistently omit the cell, the grounding is simply following the summary's omissions rather than the image content. A sharper version is to corrupt the segmentation map by moving a class's pixels to a different cell while keeping the question fixed: if Checkmate's predicted cells follow the corrupted map instead of the original image, then the explanation is only as faithful as the segmenter.

Watch

Extended reading notes

Core claim

Checkmate's central discovery is that a semantic segmentation map, reduced to a structured textual summary of per-cell class occupancy, is enough for a language-only transformer to perform RSVQA while also reporting the spatial cells used for the decision. On the new Chessboard dataset, built so that no answer dominates, with 335 possible answers, a prior bias score of 0.06 on answers and near zero on cells, Checkmate with SegFormer achieves 50.9% average accuracy on answer prediction and 82.5% micro-F1 on cell prediction, with lower inter-cell correlation than all baseline models, which the paper interprets as more focused and less redundant visual reasoning. The oracle version of the pipeline, fed ground-truth segmentation maps, reaches 94.3% average accuracy, showing that the text-summary mechanism itself can carry the task and that most of the gap in the full model is caused by segmentation errors. This is what makes the model interpretable by design: the answer and the explanation come from the same intermediate representation, not from a post hoc explainer.

Load-bearing premise

The load-bearing premise is that the text summary produced from the segmentation map preserves every piece of visual information needed to answer the question and to locate the supporting cells; if that summary drops or distorts anything relevant, the model's predicted cells are not a faithful explanation.

Editorial extensions

If this is right

  • Every prediction Checkmate returns can be inspected as a pair: the text answer plus the predicted cells, so a user can immediately see whether the model answered from the right part of the image.
  • Chessboard's balanced design removes the easy route of guessing the majority answer, so future models trained on it will have to rely on visual content to score well.
  • Because the language model receives only the summarized segmentation, improving the semantic segmenter directly improves the VQA answers; the 0.943 oracle accuracy versus 0.509 for SegFormer quantifies the available head room.
  • The cell-prediction metrics give RSVQA a second, independent evaluation axis, allowing models to be compared on whether they locate evidence rather than solely on whether they give the right string.

Reading between the lines

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

  • Editorial inference: the 30-pixel drop rule means classes occupying tiny patches are silently removed; testing thresholds at 10, 50, or 100 pixels would show how much of the area-answer weakness is caused by this summarization choice.
  • Editorial inference: the fixed 4-by-4 cell grid treats the image as a chessboard, but elongated objects such as rivers cross cells; letting the segmenter propose connected-component regions as grounding units could improve localization for such classes.
  • Editorial inference: because the language model sees only text, the model cannot correct a segmentation error; a variant that also exposes the raw image to a vision encoder would test how much the summary bottleneck actually costs.
  • Editorial inference: the answer-plus-cells output is directly usable as a human-AI collaboration interface in disaster mapping, where an operator needs to verify quickly whether the model noticed the same region a human would notice.
Share X Bluesky LinkedIn Reddit HN

Signed reviews

No signed human review yet.

Editorial analysis

A structured set of objections, weighed in public.

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

Referee Report

3 major / 5 minor

Summary. The paper introduces Chessboard, a large-scale RSVQA dataset built on reBEN with 3,123,253 image/question/answer triplets, 16-cell spatial groundings, and explicit bias-minimization through balancing. It also introduces Checkmate, a two-stage model that first produces a semantic segmentation map, converts it into a structured text summary, and then uses DistilBERT with two classification heads to predict both the answer and the relevant cells. The authors evaluate four segmentation backbones, compare Checkmate with several VQA baselines including a Blind text-only model, and report an average answer accuracy of 50.9% for Checkmate-SegFormer, a cell-prediction F1 of 0.825, and an Oracle upper bound of 0.943 average accuracy. The central claim is that the model is interpretable and explainable because the predicted cells expose the spatial locations underlying each answer.

Significance. If the main claim were fully supported, this would be a useful contribution: Chessboard is a large, intentionally balanced benchmark with explicit cell labels, and the Oracle decomposition cleanly separates the limits of the segmentation stage from the limits of the language module. The dataset's bias statistics are a concrete, falsifiable improvement over prior RSVQA datasets. However, the headline interpretability/explainability result is not yet demonstrated: the current evaluation does not show that the predicted cells are faithful to the answer computation, and the Blind model's high cell F1 indicates that the cell labels are strongly predictable from text. The paper is therefore potentially valuable but needs additional faithfulness analyses and a more carefully scoped claim.

major comments (3)
  1. [Section 4 (VQA modules) and Section 5 (Experimental Setup)] The claim that Checkmate is explainable because it identifies the image cells most relevant to its decisions is not supported by the architecture. The answer and cell predictions are produced by two separate MLPs from the same pooled DistilBERT representation, with independent losses weighted 0.6 and 0.4, and the predicted cells are never fed back into the answer module. The generated response "Based on [cells], the answer is ..." is therefore a post-hoc narrative rather than a description of the model's computation. To support the claim, the paper needs a faithfulness experiment, for example replacing predicted cells with random cells while keeping the answer head fixed and measuring answer stability, or intervening on the segmentation summary and checking that answer and cell predictions co-vary. Without such evidence, the transparency offered is a parallel cell-localization output, not an explanation of the answer computation.
  2. [Table 4] The Blind model, which receives no visual information, achieves cell F1 0.764 with recall 0.933 and correlation 0.993, while Checkmate-SegFormer reaches F1 0.825 with recall 0.780 and correlation 0.639. The six-point F1 advantage comes with lower recall, and the Blind model's high correlation suggests it may overpredict broadly. The near-parity shows that the cell labels in Chessboard are largely predictable from the question text alone, because the templates contain class names and question types that correlate with cell occurrences. The paper should quantify the marginal contribution of the visual branch, for example by reporting cell F1 per question type or by ablating the text summary (e.g., removing cell identifiers from the summary) to show that visual features are necessary for the reported cell performance. Without this, the conclusion that Checkmate provides visually grounded explanations is not established.
  3. [Sections 3 and 4 (Dataset creation and Oracle)] The Chessboard labels and the Oracle text summaries are both derived from the same reBEN segmentation maps, and the summary explicitly lists cell names next to class names. The cell-prediction task is therefore partly an entity-extraction problem over the input text, and the Oracle's cell F1 of 0.999 only shows that the summary contains enough information; it does not show that a model trained on this summary is grounding its answer in the image. The paper should either include a control condition in which the text summary is insufficient to determine the cells (for example, questions requiring comparison of area values across cells not enumerated in the summary), or restrict the interpretability claim to "cells relevant according to the generated summary." This is a scope issue rather than a claim that the dataset is useless; the dataset remains a valid benchmark for cell-conditional answer prediction, but the explanation claim needs to be bounded accordingly.
minor comments (5)
  1. [Section 3 (Question creation)] The example states "a 4-components template might be: start, placeholder, verb, preposition, and end," but the list contains five items; adjust either the count or the list.
  2. [Section 3 (Question creation)] The phrase "W ords like" contains a typo and should read "Words like".
  3. [Section 6 (Figure 3 discussion)] The statement "the mode of the distribution is one" is ambiguous; specify which distribution is meant (e.g., the distribution of area-range errors) and define the axis units in the figure or caption.
  4. [Table 2] Micro precision, recall, and F1 are reported as identical for every segmentation model; since micro precision and recall are not generally equal, please explain the aggregation or double-check the computation.
  5. [Section 5 (Baselines)] The sentence "All models use a batch size of 32 and the same loss function as the Oracle, ensuring explainability" overstates what a loss function can ensure; rephrase to state that the same loss is used and move the explainability claim to the cell-evaluation discussion.

Circularity Check

1 steps flagged · score 6.0 of 10

Cell 'explanations' are an auxiliary text-extraction head; the answer head never uses them, so the explainability claim reduces to parsing cell names already present in the input summary.

  1. fitted input called prediction [Section 3 'Cells creation'; Section 4 'VQA modules'; Section 6 'Cell Prediction']
    "The segmentation map is transformed into a textual summary through a structured template. For each cell, the algorithm identifies the present classes. ... if classA appears only in cell a1 with 458 pixels, the generated description regarding this class alone would be:T able: (a1, class_A); Area: class_A: 45001-50000m². ... The first one has an output layer of 16, corresponding to the different spatial locations in the image. ... Once the answer and the relevant cells have been predicted, a final response is constructed using a predefined template."

    The ground-truth cells are defined by the same segmentation labels that generate the text summary: Section 3 saves 'the cells in which each class appears,' and Section 4 writes those cells into the prompt as 'Table: (a1, class_A)'. So the cell target is a subset of the input tokens, and the cell head can recover it by parsing. The answer and cells come from two separate MLPs on the same pooled representation; the predicted cells are never fed back into the answer module. Thus high cell F1 (Oracle 0.999; SegFormer 0.825) measures text extraction, not a causal explanation. The final template 'Based on [cells], the answer is...' is a post-hoc overlay. The Blind model's 0.764 F1 shows the metric is not gated on visual input.

full rationale

The Chessboard dataset and answer-prediction experiments are self-contained supervised benchmarks built from reBEN labels; those parts are not circular. The circularity is confined to the explainability claim. The cell ground truth and the text summary are both generated from the same segmentation map, so the cell-classification target is literally a subset of the input text; the Oracle's near-perfect F1 is a copy/parse result, and the SegFormer F1 is a segmentation-plus-parse result. More importantly, the answer head does not consume the predicted cells, so the cells cannot be the cause of the answer. The paper's final template 'Based on [cells], the answer is...' is added after two independent predictions, making the claimed interpretability a narrative overlay rather than the model's computation. The Blind model's high cell F1 (0.764) independently shows the cell metric is not a faithful measure of visual grounding. No load-bearing self-citation or uniqueness import was found; self-citations are contextual. Because the central interpretability/explainability claim reduces to a text-extraction prediction, the paper is partially circular (score 6).

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

The central claim depends on the correctness of reBEN labels, on the sufficiency of the cell-level text summary, and on several hand-chosen thresholds (30 pixels, 5,000 m² ranges, 4x4 grid, loss weights). These parameters and assumptions are not independently validated and directly shape both the dataset and the model behavior.

free parameters (4)
  • Loss weighting for cell vs answer classification = 0.4 (cell), 0.6 (answer)
    Chosen by hand in Section 5 'VQA Modules'; this weighting shapes the final model behavior and the trade-off between the two outputs.
  • Class presence threshold = 30 pixels
    Defined in Section 3 'Cells creation'; classes occupying fewer than 30 pixels in a cell are treated as absent, directly affecting both the dataset and the text summary.
  • Area range grouping size = 5,000 m²
    Chosen in Section 3 'Balancing process' to group area values into ranges; this defines the 290 area answer classes and the granularity of the area prediction task.
  • Grid size for cells = 4x4 (16 cells)
    Chosen in Section 3 'Cells creation'; determines the spatial resolution of the explanations and the cell prediction task.
assumptions (4)
  • domain assumption reBEN segmentation labels are correct ground truth for land cover classes
    The entire dataset, the Oracle, and the segmentation training rely on reBEN's pixel-level labels (Section 3 'Dataset' and Section 4 'Oracle'). If these labels contain systematic errors, both the dataset and the model evaluation inherit them.
  • ad hoc to paper The text summary is a sufficient representation of the image for answering the questions
    The Checkmate model answers based only on the cell-level text summary (Section 4 'VQA modules'). This is a strong modeling assumption: information not captured in the summary (for example, spatial arrangement within a cell, classes below the 30-pixel threshold) is inaccessible to the VQA model.
  • ad hoc to paper The 30-pixel presence threshold preserves meaningful semantic semantics
    Defined in Section 3 'Cells creation'; the threshold is applied without analysis of its effect on answer or cell quality. A different threshold would likely change the dataset distribution and the difficulty of the cell prediction task.
  • domain assumption The balancing process does not introduce harmful selection bias
    In Section 3 'Balancing process', presence questions are balanced by taking the minimum of 'yes' and 'no' per class, and comparison answers are capped at the median frequency. This may discard informative examples and change the test distribution, but the paper does not analyze the effect.

how reviews work

0 comments
Cite this review

Pith. "Pith review of Checkmate: interpretable and explainable RSVQA is the endgame." pith.science (2026). https://pith.science/paper/MEZHXZTX

@misc{pith2026250813086,
  author       = {Pith},
  title        = {Pith review of: Checkmate: interpretable and explainable RSVQA is the endgame},
  year         = {2026},
  howpublished = {\url{https://pith.science/paper/MEZHXZTX}},
  note         = {Machine review of arXiv:2508.13086}
}
read the original abstract

Remote Sensing Visual Question Answering (RSVQA) presents unique challenges in ensuring that model decisions are both understandable and grounded in visual content. Current models often suffer from a lack of interpretability and explainability, as well as from biases in dataset distributions that lead to shortcut learning. In this work, we tackle these issues by introducing a novel RSVQA dataset, Chessboard, designed to minimize biases through 3'123'253 questions and a balanced answer distribution. Each answer is linked to one or more cells within the image, enabling fine-grained visual reasoning. Building on this dataset, we develop an explainable and interpretable model called Checkmate that identifies the image cells most relevant to its decisions. Through extensive experiments across multiple model architectures, we show that our approach improves transparency and supports more trustworthy decision-making in RSVQA systems.

Figures

Figures reproduced from arXiv: 2508.13086 by the authors.

Figure 1
Figure 1. Example of Question , Answer , and Cells in Chessboard for the four question types. The main categories are Landcover, Presence, Area, and Comparison. Within Comparison, there are two subtypes: Absolute and Relative. In Relative Comparison, the answer can be either a land cover class (LC) or a binary yes/no. using 186’731 distinct question templates and a vocabulary of 519 unique words, spanning 459’361 images from … view at source ↗
Figure 2
Figure 2. Our proposed method, Checkmate, first transforms the image into a semantic segmentation map, then summarizes it at the cell level, and finally feeds this representation into a language-only model to predict both the answer and the spatial location that influenced the decision [PITH_FULL_IMAGE:figures/full_fig_p005_2.png] view at source ↗
Figure 3
Figure 3. shows a sample image, its SegFormer generated segmentation map, and example questions with predicted and ground truth answers. Thanks to the presence of the cells in the answers, it becomes easier to understand whether the model is actually understanding the content of the image. In the first question, which involves a comparison between two classes, the answer includes cells corresponding to both classes, indicatin… view at source ↗

Review history (2 revisions) →

Discussion (0). Continue with ORCID to comment.

Reference graph

Works this paper leans on

37 extracted references · 27 canonical work pages

  1. [1]

    Visual question answering from remote sensing images

    Sylvain Lobry et al. “Visual question answering from remote sensing images”. In:IGARSS 2019-2019 IEEE International Geoscience and Remote Sensing Symposium. IEEE. 2019, pp. 4951–4954

    work page 2019

  2. [2]

    Evaluating Language Biases in Remote Sensing VQA

    Christel Chappuis et al. “Evaluating Language Biases in Remote Sensing VQA”. In:accepted for publication in the IEEE Geoscience and Remote Sensing Magazine (GRSM)(2025)

    work page 2025

  3. [3]

    Explaining explanations: An overview of interpretability of machine learning

    Leilani H Gilpin et al. “Explaining explanations: An overview of interpretability of machine learning”. In:2018 IEEE 5th International Conference on data science and advanced analytics (DSAA). IEEE. 2018, pp. 80–89

    work page 2018

  4. [4]

    How explainability contributes to trust in AI

    Andrea Ferrario and Michele Loi. “How explainability contributes to trust in AI”. In: Proceedings of the 2022 ACM conference on fairness, accountability, and transparency. 2022, pp. 1457–1466

    work page 2022

  5. [5]

    VQA: Visual Question Answering

    Stanislaw Antol et al. “VQA: Visual Question Answering”. In: IEEE/CVF ICCV. 2015, pp. 2425–2433

    work page 2015

  6. [6]

    Cross-modal visual question answering for remote sensing data: The international conference on digital image computing: Techniques and applications (DICTA 2021)

    Rafael Felix et al. “Cross-modal visual question answering for remote sensing data: The international conference on digital image computing: Techniques and applications (DICTA 2021)”. In:2021 Digital Image Computing: Techniques and Applications (DICT A). IEEE. 2021, pp. 1–9

    work page 2021

  7. [7]

    Mutual attention inception network for remote sensing visual question answering

    Xiangtao Zheng et al. “Mutual attention inception network for remote sensing visual question answering”. In:IEEE Transactions on Geoscience and Remote Sensing60 (2021), pp. 1–14

    work page 2021

  8. [8]

    Lit-4-rsvqa: Lightweight transformer-based visual question answering in remote sensing

    Leonard Hackel et al. “Lit-4-rsvqa: Lightweight transformer-based visual question answering in remote sensing”. In: IGARSS 2023-2023 IEEE International Geoscience and Remote Sensing Symposium. IEEE. 2023, pp. 2231–2234

    work page 2023

Show all 37 references
  1. [9]

    Tinybert: Distilling bert for natural language understanding

    Xiaoqi Jiao et al. “Tinybert: Distilling bert for natural language understanding”. In:arXiv preprint arXiv:1909.10351 (2019)

    2019 arXiv

  2. [10]

    Multi-modal fusion transformer for visual question answering in remote sensing

    Tim Siebert et al. “Multi-modal fusion transformer for visual question answering in remote sensing”. In:Image and Signal Processing for Remote Sensing XXVIII. V ol. 12267. SPIE. 2022, pp. 162–170

    2022

  3. [11]

    Remote sensing chatgpt: Solving remote sensing tasks with chatgpt and visual models

    Haonan Guo et al. “Remote sensing chatgpt: Solving remote sensing tasks with chatgpt and visual models”. In:IGARSS 2024-2024 IEEE International Geoscience and Remote Sensing Symposium. IEEE. 2024, pp. 11474–11478

    2024

  4. [12]

    Multilingual augmentation for robust visual question answering in remote sensing images

    Zhenghang Yuan, Lichao Mou, and Xiao Xiang Zhu. “Multilingual augmentation for robust visual question answering in remote sensing images”. In:2023 Joint Urban Remote Sensing Event (JURSE). IEEE. 2023, pp. 1–4

    2023

  5. [13]

    LLM-Driven Data Augmentation for Visual Question Answering

    Boussaid Hichem et al. “LLM-Driven Data Augmentation for Visual Question Answering”. In:2025 Joint Urban Remote Sensing Event (JURSE). IEEE. 2025, pp. 1–4

    2025

  6. [14]

    RSVQA meets BigEarthNet: a new, large-scale, visual question answering dataset for remote sensing

    Sylvain Lobry, Begüm Demir, and Devis Tuia. “RSVQA meets BigEarthNet: a new, large-scale, visual question answering dataset for remote sensing”. In:2021 IEEE International Geoscience and Remote Sensing Symposium IGARSS. IEEE. 2021, pp. 1218–1221

    2021

  7. [15]

    Floodnet: A high resolution aerial imagery dataset for post flood scene understanding

    Maryam Rahnemoonfar et al. “Floodnet: A high resolution aerial imagery dataset for post flood scene understanding”. In: IEEE Access 9 (2021), pp. 89644–89654

    2021

  8. [16]

    Stop explaining black box machine learning models for high stakes decisions and use interpretable models instead

    Cynthia Rudin. “Stop explaining black box machine learning models for high stakes decisions and use interpretable models instead”. In:Nature machine intelligence1.5 (2019), pp. 206–215

    2019

  9. [17]

    Interpretability and explainability: A machine learning zoo mini-tour

    Ričards Marcinkevičs and Julia E V ogt. “Interpretability and explainability: A machine learning zoo mini-tour”. In: arXiv preprint arXiv:2012.01805(2020)

    2020 arXiv

  10. [18]

    Prompt-RSVQA: Prompting visual context to a language model for remote sensing visual question answering

    Christel Chappuis et al. “Prompt-RSVQA: Prompting visual context to a language model for remote sensing visual question answering”. In:Proceedings of the IEEE/CVF conference on computer vision and pattern recognition. 2022, pp. 1372– 1381

    2022

  11. [19]

    SAR Strikes Back: A New Hope for RSVQA

    Lucrezia Tosato et al. “SAR Strikes Back: A New Hope for RSVQA”. In:IEEE Journal of Selected Topics in Applied Earth Observations and Remote Sensing(2025)

    2025

  12. [20]

    Segmentation-guided attention for visual question answering from remote sensing images

    Lucrezia Tosato et al. “Segmentation-guided attention for visual question answering from remote sensing images”. In: IGARSS 2024-2024 IEEE International Geoscience and Remote Sensing Symposium. IEEE. 2024, pp. 2750–2754

    2024

  13. [21]

    Hierarchical Multi-Modality Graph Reasoning for Remote Sensing Visual Question Answering

    Han Zhang et al. “Hierarchical Multi-Modality Graph Reasoning for Remote Sensing Visual Question Answering”. In: IEEE Transactions on Geoscience and Remote Sensing(2024)

    2024

  14. [22]

    PAN-RSVQA: Vision Foundation Models as Pseudo-ANnotators for Remote Sensing Visual Question Answering

    Christel Chappuis et al. “PAN-RSVQA: Vision Foundation Models as Pseudo-ANnotators for Remote Sensing Visual Question Answering”. en. In:Proceedings of the 2025 IEEE/CVF Conference on Computer Vision and Pattern Recognition W orkshops (CVPRW). Nashville, TN, USA: IEEE, 2025

    2025

  15. [23]

    Rsadapter: Adapting multimodal models for remote sensing visual question answering

    Yuduo Wang and Pedram Ghamisi. “Rsadapter: Adapting multimodal models for remote sensing visual question answering”. In:IEEE Transactions on Geoscience and Remote Sensing(2024)

    2024

  16. [24]

    GEMeX: A Large-Scale, Groundable, and Explainable Medical VQA Benchmark for Chest X-ray Diagnosis

    Bo Liu et al. “GEMeX: A Large-Scale, Groundable, and Explainable Medical VQA Benchmark for Chest X-ray Diagnosis”. In: arXiv preprint arXiv:2411.16778(2024). Checkmate: interpretable and explainable RSVQA is the endgame 11

    2024 arXiv

  17. [25]

    reben: Refined bigearthnet dataset for remote sensing image analysis

    Kai Norman Clasen et al. “reben: Refined bigearthnet dataset for remote sensing image analysis”. In:arXiv preprint arXiv:2407.03653 (2024)

    2024 arXiv

  18. [26]

    Bigearthnet: A large-scale benchmark archive for remote sensing image understanding

    Gencer Sumbul et al. “Bigearthnet: A large-scale benchmark archive for remote sensing image understanding”. In: IGARSS 2019-2019 IEEE international geoscience and remote sensing symposium. IEEE. 2019, pp. 5901–5904

    2019

  19. [27]

    U-NET: Convolutional networks for biomedical image segmentation

    Olaf Ronneberger, Philipp Fischer, and Thomas Brox. “U-NET: Convolutional networks for biomedical image segmentation”. In: Medical image computing and computer-assisted intervention–MICCAI 2015: 18th international conference, Munich, Germany, October 5-9, 2015, proceedings, pa...

    2015

  20. [28]

    SegFormer: Simple and efficient design for semantic segmentation with transformers

    Enze Xie et al. “SegFormer: Simple and efficient design for semantic segmentation with transformers”. In:Advances in neural information processing systems34 (2021), pp. 12077–12090

    2021

  21. [29]

    Neural plasticity-inspired foundation model for observing the Earth crossing modalities

    Zhitong Xiong et al. “Neural plasticity-inspired foundation model for observing the Earth crossing modalities”. In: arXiv e-prints (2024), arXiv–2403

    2024

  22. [30]

    DistilBERT, a distilled version of BERT: smaller, faster, cheaper and lighter

    Victor Sanh et al. “DistilBERT, a distilled version of BERT: smaller, faster, cheaper and lighter”. In: arXiv preprint arXiv:1910.01108 (2019)

    2019 arXiv

  23. [31]

    Visualbert: A simple and performant baseline for vision and language

    Liunian Harold Li et al. “Visualbert: A simple and performant baseline for vision and language”. In: arXiv preprint arXiv:1908.03557 (2019)

    2019 arXiv

  24. [32]

    Vl-bert: Pre-training of generic visual-linguistic representations

    Weijie Su et al. “Vl-bert: Pre-training of generic visual-linguistic representations”. In:arXiv preprint arXiv:1908.08530 (2019)

    2019 arXiv

  25. [33]

    PERS: Parameter-Efficient Multi-modal Transfer Learning for Remote Sensing Visual Question Answering

    Jinlong He et al. “PERS: Parameter-Efficient Multi-modal Transfer Learning for Remote Sensing Visual Question Answering”. In:IEEE Journal of Selected Topics in Applied Earth Observations and Remote Sensing(2024)

    2024

  26. [34]

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

    Alexey Dosovitskiy et al. “An image is worth 16x16 words: Transformers for image recognition at scale”. In:arXiv preprint arXiv:2010.11929 (2020)

    2020 arXiv

  27. [35]

    BERT: Pre-training of Deep Bidirectional Transformers for Language Understanding

    Jacob Devlin et al. “BERT: Pre-training of Deep Bidirectional Transformers for Language Understanding”. In:Proceedings of the 2019 Conference of the North American Chapter of the Association for Computational Linguistics: Human Language Technologies, V olume 1 (Long and Short ...

    2019 doi

  28. [36]

    Visual Question Answering on Multiple Remote Sensing Image Modalities

    Hichem Boussaid et al. “Visual Question Answering on Multiple Remote Sensing Image Modalities”. en. In:Proceedings of the 2025 IEEE/CVF Conference on Computer Vision and Pattern Recognition W orkshops (CVPRW). Nashville, TN, USA: IEEE, 2025

    2025

  29. [37]

    OpenEarthMap-SAR: A Benchmark Synthetic Aperture Radar Dataset for Global High-Resolution Land Cover Mapping

    Junshi Xia et al. “OpenEarthMap-SAR: A Benchmark Synthetic Aperture Radar Dataset for Global High-Resolution Land Cover Mapping”. In:arXiv preprint arXiv:2501.10891(2025)

    2025 arXiv

Pith tools

Reviewed August 15, 2026 · model on record in the stance chip above.