Pith. sign in

REVIEW 4 major objections 7 minor 21 references

Injecting detector-derived spatial predicates into prompts, with depth captions, fixes multimodal models’ fine-grained spatial reasoning without any training.

Reviewed by Pith at T0; open to challenge. T0 means a machine referee read the full paper against a public rubric. the ladder, T0–T4 →

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

Training-free injection of YOLO-World pairwise geometric predicates plus depth captions raises MLLM spatial F1 (notably +46% relative on BLINK for Qwen2.5-VL) and recovers weak models on VSR.

T0 review reviewed 2026-07-30 challenge →

load-bearing objection Sensible training-free hybrid of LDP depth captions and YOLO box predicates; useful engineering, but the marquee BLINK F1 numbers do not match the paper's own F1 formula and gains are mixed for the strongest model. the 4 major comments →

arxiv 2607.27145 v1 pith:4DULKM3H submitted 2026-07-29 cs.CV

Explainable and Resource-Efficient Spatial Reasoning in Multimodal LLMs for Decision-Critical Applications

Piyush Jain , Kousik Dasgupta , Rajarshi Roy , Subarna Tripathi This is my paper
classification cs.CV
keywords Multimodal Large Language ModelsSpatial ReasoningExplainable AIDecision Support SystemsHallucination ReductionResource-Efficient InferenceLayered-Depth-Based PromptingOpen-Vocabulary Detection
verification ladder T0 review T1 audit T2 compute T3 formal T4 reserved

The pith

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

The reading

Multimodal language models are moving into robotics, embodied agents, and safety monitoring, where a wrong left/right or inside judgment can matter and where black-box answers are hard to audit. Depth-layer prompting already helps with near-versus-far structure, but it does not tell the model how objects in the same depth band sit relative to one another. This paper shows that an open-vocabulary detector’s bounding boxes can be turned into plain-English pairwise predicates—left of, above, inside, touching, near—and injected beside the depth captions as secondary, human-readable evidence. Across spatial and hallucination benchmarks the hybrid prompt raises F1 sharply for both strong and compact models, and a predicate-only stream still fits a 40-token CPU budget. The result is a training-free scaffold that improves answers and leaves an inspectable trail for decision support.

Core claim

ByDeWay-V2 establishes that pairing layered monocular-depth captions with explicit pairwise geometric predicates computed from open-vocabulary bounding boxes, then serialising both into the MLLM prompt, recovers fine-grained spatial reasoning and reduces object hallucination in a fully training-free way. On the BLINK spatial subset the method delivers a 46 percent relative F1 gain for Qwen2.5-VL over depth-only prompting; on VSR it lifts BLIP-Base from near-random F1 0.05 to 0.53; POPE metrics also improve. The same light predicate stream remains usable under a strict 40-token context limit.

What carries the argument

The SpatialAnalyser: YOLO-World detections supply axis-aligned boxes; pairwise heuristics (centre comparisons for left/right/above/below, overlap ratios for inside/touching, normalised centre distance for near/far) are serialised as readable predicates and concatenated with LDP depth-layer captions into one structured prompt.

Load-bearing premise

The method assumes that 2D box comparisons from a query-seeded detector are accurate and complete enough to serve as trustworthy secondary evidence for the model’s spatial answers.

What would settle it

Re-run BLINK spatial and VSR after systematically dropping, swapping, or corrupting the injected predicates (or replacing the detector with a much weaker one) and check whether the reported F1 gains over depth-only prompting vanish; if gains survive when the predicates are wrong, the geometric scaffold is not what is carrying the improvement.

Watch this falsifier. Get emailed when new claim-graph text bears on it.

If this is right

  • Operators can treat the injected predicates as auditable intermediate evidence rather than accepting a black-box spatial verdict.
  • Compact multimodal models that previously failed spatial VQA become competitive without fine-tuning or extra parameters.
  • Token-constrained deployments can drop verbose depth captions and keep only the compact spatial predicates.
  • Object-existence hallucination rates fall when spatial grounding is added on top of depth context alone.
  • The pipeline is model-agnostic and can sit in front of any MLLM that accepts image plus text context.

Where Pith is reading between the lines

These are editorial extensions of the paper, not claims the author makes directly.

  • Confidence-weighted or ensemble predicate injection could ease the precision–recall trade-off that appears when a strong model overrides correct context.
  • The same geometric serialisation could ground multi-step embodied instructions (move left of X, then above Y) without retraining planners.
  • Heavy occlusion and non-axis-aligned objects remain an open stress test before the method is trusted in real safety pipelines.
Share X Bluesky LinkedIn Reddit HN

Editorial analysis

A structured set of objections, weighed in public.

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

Referee Report

4 major / 7 minor

Summary. The manuscript presents ByDeWay-V2, a training-free extension of the authors' earlier Layered-Depth-Based Prompting (LDP) framework. A "SpatialAnalyser" module runs YOLO-World-L open-vocabulary detection, computes pairwise geometric relations from axis-aligned bounding boxes (above/below/left/right via centre comparisons; inside/touching via overlap ratios with thresholds θ_in=0.70, θ_touch=0.10; near/far via normalised centre distance with τ_near=0.30), serialises them as plain-English predicates, and concatenates them with LDP depth-layer captions into the MLLM prompt. Evaluation on VSR, the BLINK spatial subset, and a 150-sample curated POPE variant across Qwen2.5-VL-7B, BLIP-Base, and ViLT-B/32 reports: (i) a headline 46% relative F1 gain for Qwen2.5-VL on BLINK (0.496→0.727); (ii) recovery of BLIP-Base on VSR from a near-degenerate F1 of 0.053 to 0.525; (iii) modest POPE gains for all three models; and (iv) regressions for Qwen2.5-VL on VSR (F1 0.7835→0.7179) and for ViLT on VSR and BLINK. The method is simple, model-agnostic, and produces human-readable intermediate evidence, which is genuinely useful for auditability.

Significance. If the numbers hold, the contribution is a practical, reproducible-in-principle recipe: explicit detector-derived spatial predicates injected into prompts, with all thresholds and the full pipeline (Algorithm 1, Eqs. 3–8) disclosed, no training, and a demonstrated operating point under a 40-token budget for ViLT. The explicit LDP vs. LDP+Spatial ablation across three models and three benchmarks is a strength, as is the per-category VSR breakdown (Table 4), which localises gains to the relation types the predicates directly encode. The auditable-predicate framing is well matched to the decision-critical motivation. However, the significance is currently capped by a verification problem: the headline BLINK result (Table 3) is the only table whose F1 column is not the harmonic mean of its own precision/recall columns as defined in Eq. (12), so the paper's marquee number cannot presently be reproduced from the paper's own reported quantities. The honest reporting of the Qwen VSR regression is commendable, but the abstract's framing leans on the one benchmark/metric combination whose provenance is unclear.

major comments (4)
  1. [§4.5, Table 3 (and Eq. 12)] Table 3 is internally inconsistent with the paper's own metric definition. Applying Eq. (12) to the reported precision and recall: Qwen LDP gives 2·0.550·0.510/(0.550+0.510)=0.529, not 0.496; Qwen V2 gives 0.737, not 0.727; BLIP LDP gives 0.377, not 0.244 (a 0.13 gap); BLIP V2 gives 0.323, not 0.254; ViLT LDP gives 0.454, not 0.418; ViLT V2 gives 0.340, not 0.318. Every row of Tables 1 and 2 satisfies Eq. (12) to three decimals, so Table 3 alone was computed under some other, undisclosed protocol (e.g., macro-averaged F1 over answer classes while P/R are binary on a positive class). This is load-bearing: Table 3 is the sole quantitative support for the abstract's headline '46% relative F1 improvement'. The authors must either correct the table or fully disclose the alternative computation, including how binary TP/FP/FN are defined on BLINK, which is a multiple-choice benchmark and does n
  2. [§4.5, BLINK paragraph vs. Table 3] The text states 'BLIP-Base and ViLT also record meaningful gains in F1 and accuracy' on BLINK. The table contradicts this: ViLT F1 falls 0.418→0.318 (a substantial regression) and BLIP F1 moves only 0.244→0.254; BLIP accuracy 0.483→0.490 is within noise for a small subset. The sentence must be rewritten to match the data; as written it misrepresents two of the three model rows in the paper's headline table.
  3. [§4.5, VSR paragraph / Table 1; Abstract] For the flagship model (Qwen2.5-VL), V2 strictly regresses on VSR: F1 0.7835→0.7179, recall 0.7155→0.6130, accuracy 0.7872→0.7408. The paper's counterweight is a precision 'gain' of 0.8658→0.8660 (+0.0002), which is numerically indistinguishable from zero and cannot support the claim that the model is 'strictly more often correct'. Combined with the BLINK accuracy drop (0.748→0.727), the accurate summary for the strongest model is: V2 helps on one benchmark under one metric (whose computation is currently unexplained, see comment on Table 3) and hurts elsewhere. The abstract and §4.7 should state this trade-off plainly rather than framing a +0.0002 precision delta as a desirable property for high-stakes settings.
  4. [§3.2 and §4.1] The entire method rests on the correctness of the injected predicates, yet predicate quality is never measured. The heuristics are fragile in known ways: axis-aligned box centres give wrong left/right labels under perspective and for elongated or occluded objects; the priority order containment > contact > vertical/horizontal > proximity forces exactly one predicate per ordered pair even when several hold; and the detector vocabulary is seeded from the query nouns at conf≥0.20, so missed detections silently remove the relevant evidence. §4.7 acknowledges detector errors anecdotally ('earlier experiments') but reports no failure analysis. A small audit — e.g., predicate accuracy against VSR ground-truth relations on a few hundred samples, and performance stratified by whether the queried pair was actually detected — would directly test the paper's central assumption and is within the manu
minor comments (7)
  1. [§4.2, POPE] POPE evaluation uses a 150-sample curated subset hosted by an author (ref. 14) rather than the standard POPE splits. With n=150, the reported accuracy differences (e.g., 86.0%→90.7%) have wide confidence intervals; please report CIs or evaluate on the full benchmark, and state the curation criteria for the subset.
  2. [§4.5 / §4.6] No variance or significance reporting anywhere; all numbers appear to be single runs. For the large VSR test set this is less concerning, but the BLINK spatial subset is small, and the Qwen accuracy drop there (0.748→0.727) may not be significant. Please give subset sizes and, where feasible, bootstrap CIs.
  3. [§3.2 / §4.1] §3.2 says all N(N−1) ordered-pair sentences are concatenated into C_Spatial, while §4.1 caps at 20 spatial relation pairs per image. Please specify the selection/pruning rule when N(N−1) exceeds the cap, since it affects which evidence the model sees.
  4. [§4.5] The 'Vision-First Advisory' instruction is credited with the Qwen VSR regression but is never shown verbatim; only Figure 3 paraphrases it ('Trust your visual analysis FIRST'). Include the exact prompt text (appendix is fine), since the ablation LDP vs. LDP+Spatial is confounded if the instruction wording also changed between V1 and V2 — please confirm the instruction was identical across both configurations.
  5. [General] No code or prompt release is mentioned. Given that the contribution is a prompt-construction pipeline with several free thresholds (θ_in, θ_touch, τ_near, conf≥0.20, max 20 pairs, T1/T2 percentiles), releasing the serialisation code and exact prompts would substantially raise reproducibility.
  6. [§4.3 / Abstract] The abstract's 'operates under a strict 40-token context budget on CPU' conflates two properties: the 40-token limit applies to ViLT (a 40-token text encoder limit, not a context budget in the usual LLM sense), and CPU inference is mentioned only for ViLT in §4.1. Please clarify which model/configuration supports the real-time CPU claim and report latency.
  7. [§4.6, Table 4] All seven VSR categories show near-uniform ΔF1 (+0.39 to +0.57) including 'Unallocated', which by definition lacks a clear relation type. A brief comment on why unallocated relations benefit as much as projective ones would help rule out a generic answer-distribution shift (e.g., V2 simply making BLIP answer 'true' more often) rather than relation-specific grounding.

Circularity Check

0 steps flagged

No circularity: empirical prompt-injection method evaluated on external benchmarks; results are not forced by construction or self-citation.

full rationale

ByDeWay-V2 is a training-free prompting pipeline: monocular depth captions (LDP) plus pairwise geometric predicates from YOLO-World-L bounding boxes are concatenated into the MLLM context and scored against external labels on VSR, BLINK, and POPE. The geometric rules (center comparisons, fixed OR/distance thresholds) are independent of those labels; they do not fit parameters on the test relations and then re-predict them. Citing the authors' prior ByDeWay/LDP work and a small POPE subset they hosted is ordinary baseline/self-reference, not a load-bearing uniqueness theorem or a definition that makes the reported F1 gains tautological. The paper claims empirical accuracy/F1 lifts, not a first-principles derivation that reduces to its inputs. Any Table-3 F1/metric inconsistencies are correctness issues, not circularity. Score 0; steps empty.

Axiom & Free-Parameter Ledger

5 free parameters · 4 axioms · 1 invented entities

The central empirical claim rests on standard CV components (monocular depth, open-vocab detection, box geometry) plus several hand-chosen thresholds and the assumption that 2D box predicates are adequate proxies for the linguistic spatial relations in VSR/BLINK. No new physical entities; free parameters are the geometric cutoffs and detection settings.

free parameters (5)
  • θ_in (inside overlap ratio) = 0.70
    Hand-set threshold OR>0.70 declares ‘inside’; directly controls topological predicates fed to the MLLM.
  • θ_touch (touching overlap ratio) = 0.10
    Hand-set band 0.10<OR≤0.70 declares ‘touching’.
  • τ_near (proximity distance) = 0.30
    Normalised center distance <0.30 declares ‘near’.
  • YOLO confidence threshold and max pairs = 0.20 conf, max 20 pairs
    Detection conf≥0.20 and cap of 20 relation pairs per image shape which predicates appear.
  • Depth percentile cuts T1/T2 = 30th and 70th percentiles
    30th/70th percentiles define close/mid/far LDP layers inherited from V1.
axioms (4)
  • domain assumption Axis-aligned 2D bounding-box geometry (centers, IoU-style overlap, normalised Euclidean distance) is a sufficient proxy for projective and topological relations queried in VSR/BLINK.
    Invoked throughout §3.2 relation computation; failures under occlusion or 3D pose are not modelled.
  • domain assumption YOLO-World-L open-vocabulary detections seeded by query entities plus COCO fallback adequately cover the objects needed for the spatial query.
    Stage 2 detection (§3.2, §4.1); missed detections silently omit predicates.
  • ad hoc to paper Priority order containment > contact > vertical/horizontal > proximity yields a unique readable predicate per ordered pair.
    Stated in §3.2; alternative priority or multi-label serialisation not compared.
  • domain assumption MLLMs will treat injected natural-language predicates as useful secondary evidence under the stated ‘Vision-First’ instruction without catastrophic over-reliance or under-reliance.
    Prompt design in Fig. 3 and discussion §4.7; Qwen VSR recall drop shows the tension.
invented entities (1)
  • SpatialAnalyser module (pairwise geometric predicate serialiser) no independent evidence
    purpose: Turns detector boxes into English spatial sentences concatenated with LDP captions.
    Named pipeline component in §3.2; engineering module rather than a new scientific entity; no independent existence outside the prompt construction.

reviewed 2026-07-30 · how reviews work

0 comments
Cite this review

Pith. "Pith review of Explainable and Resource-Efficient Spatial Reasoning in Multimodal LLMs for Decision-Critical Applications." pith.science (2026). https://pith.science/paper/4DULKM3H

@misc{pith2026260727145,
  author       = {Pith},
  title        = {Pith review of: Explainable and Resource-Efficient Spatial Reasoning in Multimodal LLMs for Decision-Critical Applications},
  year         = {2026},
  howpublished = {\url{https://pith.science/paper/4DULKM3H}},
  note         = {Machine review of arXiv:2607.27145}
}
Share X Bluesky LinkedIn Reddit HN
read the original abstract

As Multimodal Large Language Models (MLLMs) are increasingly deployed in decision-critical pipelines such as robotics, embodied AI, and safety monitoring, the opacity of their spatial judgments limits operator trust and auditability. MLLMs demonstrate strong reasoning but often struggle with fine-grained spatial understanding and object hallucination. Prior work, ByDeWay, introduced Layered-Depth-Based Prompting (LDP), a training-free framework that mitigates hallucinations by structuring prompts using monocular depth estimation. However, coarse depth layering falls short in resolving object-to-object spatial relationships within the same geometric plane, such as projective ("left of", "above") and topological ("inside", "touching") relations. We propose ByDeWay-V2, which integrates explicit spatial relational context alongside depth cues, expressed as human-readable predicates that serve as auditable evidence for downstream decision support. Using an open-vocabulary object detector (YOLO-World-L), our framework computes pairwise geometric relations between detected objects and injects them as structured spatial predicates into the MLLM prompt, bridging 3D scene depth and 2D spatial semantics without any training. We evaluate ByDeWay-V2 on the Visual Spatial Reasoning (VSR) and BLINK benchmarks across multiple MLLMs, with hallucination grounding assessed via POPE. On the BLINK spatial subset, ByDeWay-V2 achieves a 46 percent relative F1 improvement over LDP for Qwen2.5-VL, and recovers BLIP-Base's spatial reasoning on VSR from near-random performance to a competitive F1 of 0.53. Our lightest configuration operates under a strict 40-token context budget on CPU, showing the framework's suitability for resource-constrained, real-time decision-support settings.

Figures

Figures reproduced from arXiv: 2607.27145 by Kousik Dasgupta, Piyush Jain, Rajarshi Roy, Subarna Tripathi.

Figure 1
Figure 1. Figure 1: ByDeWay-V2 Workflow. The framework runs two parallel branches on the in￾put image. (Top branch) YOLO-World-L detects all objects and returns precise bounding boxes; pairwise geometric heuristics then derive explicit spatial predicates (e.g. “laptop is left of person 1”, “person 1 is above laptop”). (Bottom branch) DepthAnything V2 produces a dense depth map that is partitioned into three layers (closest, m… view at source ↗
Figure 2
Figure 2. Figure 2: YOLO-World-L open-vocabulary detection on a high-resolution sample. The model successfully isolates the two target entities specified in a spatial query (“The cup is at the left side of the laptop.”): laptop (conf. 0.98) and cup (conf. 0.96). These unambiguous, high-confidence bounding boxes serve as the geometric primitives from which the SpatialAnalyser module computes pairwise spatial predicates (e.g. l… view at source ↗
Figure 3
Figure 3. Figure 3: Prompt comparison for spatial query: “The cup is at the left side of the laptop.” (Top) The baseline prompt provides no spatial grounding. (Bottom) The ByDeWay-V2 prompt supplies LDP layer captions and explicit YOLO-World-L spatial predicates extracted at inference time. The computed predicate (cup is left of laptop) directly aligns with the query, allowing the MLLM to confidently answer true. 4 Experiment… view at source ↗

discussion (0)

Sign in with ORCID, Apple, or X to comment. Anyone can read and Pith papers without signing in.

Reference graph

Works this paper leans on

21 extracted references · 6 linked inside Pith

  1. [1]

    arXiv preprint arXiv:2502.13923 (2025)

    Bai, S., Chen, K., Liu, X., Wang, J., Ge, W., Song, J., Wang, P., Luo, K., Wang, K., Li, H., et al.: Qwen2.5-vl technical report. arXiv preprint arXiv:2502.13923 (2025)

    Pith/arXiv arXiv 2025

  2. [2]

    arXiv preprint arXiv:2404.18930 (2024)

    Bai, Z., Wang, P., Xiao, T., He, T., Han, Z., Zhang, Z., Shou, M.Z.: Hallucination of multimodal large language models: A survey. arXiv preprint arXiv:2404.18930 (2024)

    Pith/arXiv arXiv 2024

  3. [3]

    In: Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition (CVPR)

    Chen, B., Xu, Z., Kirmani, S., Ichter, B., Sadigh, D., Guibas, L., Xia, F.: Spatialvlm: Endowing vision-language models with spatial reasoning capabilities. In: Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition (CVPR). pp. 14455–14465 (June 2024)

    2024

  4. [4]

    In: Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition (CVPR) (2021)

    Chen, D.Z., Gholami, A., Nießner, M., Chang, A.X.: Scan2cap: Context-aware dense captioning in rgb-d scans. In: Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition (CVPR) (2021)

    2021

  5. [5]

    In: Proc

    Cheng, T., Song, L., Ge, Y., Liu, W., Wang, X., Shan, Y.: Yolo-world: Real-time open-vocabulary object detection. In: Proc. IEEE Conf. Computer Vision and Pattern Recognition (CVPR) (2024)

    2024

  6. [6]

    arXiv preprint arXiv:2404.12390 (2024)

    Fu, X., Hu, Y., Li, B., Feng, Y., Wang, H., Lin, X., Roth, D., Smith, N.A., Ma, W.C., Krishna, R.: Blink: Multimodal large language models can see but not perceive. arXiv preprint arXiv:2404.12390 (2024)

    Pith/arXiv arXiv 2024

  7. [7]

    In: Proceedings of the 38th International Conference on Machine Learning (2021)

    Kim, W., Son, B., Kim, I.: Vilt: Vision-and-language transformer without convolu- tion or region supervision. In: Proceedings of the 38th International Conference on Machine Learning (2021)

    2021

  8. [8]

    In: ICML (2023)

    Li, J., Li, D., Savarese, S., Hoi, S.: Blip-2: Bootstrapping language-image pre-training with frozen image encoders and large language models. In: ICML (2023)

    2023

  9. [9]

    In: Proceedings of EMNLP (2023)

    Li, Y., Du, Y., Zhou, K., Wang, J., Zhao, W.X., Wen, J.R.: Evaluating object hallucination in large vision-language models. In: Proceedings of EMNLP (2023)

    2023

  10. [10]

    Transactions of the Association for Computational Linguistics (2023)

    Liu, F., Emerson, G.E.T., Collier, N.: Visual spatial reasoning. Transactions of the Association for Computational Linguistics (2023)

    2023

  11. [11]

    Liu, H., Li, C., Wu, Q., Lee, Y.J.: Visual instruction tuning (2023)

    2023

  12. [12]

    arXiv preprint arXiv:2303.05499 (2023)

    Liu, S., Zeng, Z., Ren, T., et al.: Grounding dino: Marrying dino with grounded pre-training for open-set object detection. arXiv preprint arXiv:2303.05499 (2023)

    Pith/arXiv arXiv 2023

  13. [13]

    arXiv preprint arXiv:2306.14824 (2023) 14 Piyush Jain, Kousik Dasgupta, Rajarshi Roy, and Subarna Tripathi

    Peng, Z., Wang, W., Dong, L., Hao, Y., Huang, S., Ma, S., Wei, F.: Kosmos- 2: Grounding multimodal large language models to the world. arXiv preprint arXiv:2306.14824 (2023) 14 Piyush Jain, Kousik Dasgupta, Rajarshi Roy, and Subarna Tripathi

    Pith/arXiv arXiv 2023

  14. [14]

    https://huggingface.co/datasets/Ra jarshi-Roy-research/lmms-lab-POPE-depth-captions(2024)

    Roy, R.: lmms-lab-pope-depth-captions. https://huggingface.co/datasets/Ra jarshi-Roy-research/lmms-lab-POPE-depth-captions(2024)

    2024

  15. [15]

    Roy, R., Das, D., Banerjee, A., Bhattacharjee, A., Dasgupta, K., Tripathi, S.: Bydeway: Boost your multimodal llm with depth prompting in a training-free way (2025)

    2025

  16. [16]

    Sun, Y., Hao, J., Zhu, K., Liu, J.J., et al.: Descriptive caption enhancement with visual specialists for multimodal perception (2025)

    2025

  17. [17]

    In: Proceedings of the International Joint Conference on Artificial Intelligence (IJCAI) (2022)

    Wang, H., Zhang, C., Yu, J., Cai, W.: Spatiality-guided transformer for 3d dense captioning on point clouds. In: Proceedings of the International Joint Conference on Artificial Intelligence (IJCAI) (2022)

    2022

  18. [18]

    In: NeurIPS (2022)

    Wei, J., Wang, X., Schuurmans, D., Bosma, M., Ichter, B., Xia, F., Chi, E., Le, Q., Zhou, D.: Chain-of-thought prompting elicits reasoning in large language models. In: NeurIPS (2022)

    2022

  19. [19]

    Pattern Recognition164, 111522 (2025)

    Xu, D., Huang, Q., Zhang, X., Cheng, H., Shuang, F., Cai, Y.: Device: Depth and visual concepts aware transformer for ocr-based image captioning. Pattern Recognition164, 111522 (2025)

    2025

  20. [20]

    Yang, J., Zhang, H., Li, F., Zou, X., Li, C., Gao, J.: Set-of-mark prompting unleashes extraordinary visual grounding in gpt-4v (2023)

    2023

  21. [21]

    arXiv preprint arXiv:2406.09414 (2024)

    Yang, L., Kang, B., Huang, Z., Zhao, Z., Xu, X., Feng, J., Zhao, H.: Depth anything v2. arXiv preprint arXiv:2406.09414 (2024)

    Pith/arXiv arXiv 2024

This paper was first reviewed by grok-4.5 on July 30, 2026.