Pith. sign in

REVIEW 5 major objections 5 minor 28 references

Exploring Primitive Visual Measurement Understanding and the Role of Output Format in Learning in Vision-Language Models

T0 review · 5 major / 5 minor · reviewed 2026-08-10 · deepseek-v4-flash

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

Pith's one-line read This paper claims that how a vision-language model is told to write its answer changes how well it measures what it sees, with sentence-formatted targets beating tuple-formatted ones and numeric-token loss scaling further improving…

desk verdict A useful new benchmark and a plausible training recipe, but the sentence-vs-tuple headline is overstated and the main comparison may be partly an artifact of the evaluation parser. read the letter →

arxiv 2501.15144 v2 pith:BXKY2WHS submitted 2025-01-25 cs.CV

Ankit Yadav , Lingqiao Liu , Yuankai Qi This is my paper · ORCID
classification cs.CV
keywords vision-languagemodelsout-of-domaingeneralizationoutputformatspatialmeasurementlossscalingprimitiveshapesstructuredattributematchingaccuracyfine-tuning
verification ladder T0 review T1 audit T2 compute T3 formal

The pith

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

The reading

This paper asks whether the way a vision-language model is told to write its answer changes how well it can measure what it sees. By fine-tuning open-source vision-language models of 2B to 8B parameters on a controlled benchmark of simple 2D shapes, the authors try to show that coherent sentence-style outputs produce better spatial and numerical predictions than compact tuple-style outputs, especially when test conditions move far from training. They also claim that scaling the loss on numeric tokens during fine-tuning improves coordinate and rotation estimation. The paper introduces its own benchmark and a matching-based accuracy metric, and validates the format effect on a real plant-phenotyping dataset under domain shift. If these claims hold, output-format design and loss weighting become cheap, concrete levers for improving fine-grained spatial reasoning in vision-language models.

What carries the argument

The argument is carried by four coupled devices. The benchmark generates images from known configurations of shapes, colors, quadrants, center coordinates, rotations, occlusions, and relative positions, so every prediction has a ground truth. The two output formats, sentence and tuple, encode the same ground truth differently, isolating the effect of surface format on learning. The evaluation metric, Structured Attribute Matching Accuracy (SAMA), uses the Jonker-Volgenant algorithm to solve the linear assignment between predicted and ground-truth shape-attribute sets, with edit distance as the cost, then reports per-attribute accuracy and RMSE for continuous attributes. The fourth device is loss scaling: multiplying the cross-entropy loss on predefined numeric tokens during fine-tuning, which the paper shows improves numeric approximation.

What would settle it

Run the sentence-versus-tuple comparison with a format-agnostic scorer that converts both output formats into the same structured object before matching, and check whether the sentence advantage in SAMA and RMSE survives; if it shrinks or reverses, the claim that sentence format improves learning would not be supported.

Watch

Extended reading notes

Core claim

The paper's central claim is that output format is a learning signal for vision-language models, not just a reporting choice. On a synthetic benchmark of 2D shapes with controlled attributes, models fine-tuned on full-sentence descriptions of each shape consistently beat models fine-tuned on tuple-formatted outputs carrying the same information, and the advantage grows on out-of-domain test sets with larger domain gaps. The paper further claims that scaling the cross-entropy loss on numeric tokens by a factor of roughly 2 to 2.5 improves numeric approximation, lowering coordinate error further at a small accuracy cost. A real-world plant-phenotyping experiment supports the format effect: under the hardest distribution shift, leaf-count error is substantially lower for sentence-format fine-tuning than for tuple-format fine-tuning. The paper also introduces SAMA, a Jonker-Volgenant assignment-based metric, to score shape-attribute predictions when predicted shapes appear in arbitrary order.

Load-bearing premise

The results depend on the assumption that the evaluation pipeline scores sentence and tuple outputs fairly, but SAMA extracts attributes with pattern matching that may parse one format more easily than the other.

Editorial extensions

If this is right

  • Sentence-formatted targets become the safer default for fine-tuning VLMs on structured spatial tasks, since the paper's experiments show they improve both attribute-matching accuracy and coordinate and rotation error across out-of-domain sets.
  • Weighting numeric tokens in the loss by a factor of 2 to 2.5 is a cheap, complementary intervention that lowers coordinate RMSE, with a small accuracy trade-off that can be tuned.
  • The format advantage widens as distribution shift increases, so output-format choice matters most exactly when generalization is hardest.
  • Larger models gain more from sentence format than smaller models, which can detect attributes but often misassign them; format interventions alone are unlikely to fix small-model spatial grouping.
  • SAMA offers a reusable way to score compositional outputs with unordered shape-attribute sets, which could transfer to other structured prediction benchmarks.

Reading between the lines

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

  • The authors do not test this, but the format effect suggests that aligning output structure with the LLM's pretraining distribution is a general training design choice, applicable to bounding-box regression, depth estimation, and other continuous-output VLM tasks.
  • A testable extension: paraphrase the same sentence template to check whether the advantage comes from surface form or from coherent sentence structure.
  • Another extension: combine sentence format with numeric-token loss scaling on larger models and measure whether the gains are additive.
  • The synthetic benchmark could be pushed with unseen colors, shape combinations, or perspective distortion to see where the format advantage saturates.
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 / 5 minor

Summary. The paper introduces a synthetic benchmark of 2D primitive shapes with varied spatial attributes, fine-tunes six open-source VLMs (2B-8B) with LoRA, and studies two output formats (sentence vs. tuple) plus a numeric-token loss-scaling technique. It proposes a custom metric, SAMA, based on Jonker-Volgenant matching and regex-based attribute extraction, and validates the approach on a plant-phenotyping dataset. The headline claims are that sentence-formatted targets outperform tuple-formatted targets, especially under large out-of-domain shifts, and that scaling the loss on numeric tokens improves numerical approximation. The code is publicly released.

Significance. If the claims held, the paper would provide a simple, practical recipe for fine-tuning VLMs on spatial measurement tasks: use sentence-formatted targets and weight numeric-token losses. The benchmark itself, with controlled OD scenarios and multiple model sizes, is a useful contribution, and the inclusion of a real-world plant-phenotyping transfer experiment strengthens the empirical scope. The paper is also generally reproducible in structure: hyperparameters, LoRA ranks, and dataset construction are described in detail, and the code is available. However, the headline claims are broader than the reported data support, and the main comparison is vulnerable to a parser confound and to the absence of uncertainty estimates, so the significance of the conclusions is currently limited.

major comments (5)
  1. [Section 4.2, Tables 2, 3, and 6] The claim that 'models fine-tuned on the sentence format consistently outperform their tuple-based counterparts' and the abstract's statement that 'sentence-based outputs outperform tuple formats' are contradicted by the paper's own data for smaller models. In Table 2, Qwen-VL 2B achieves higher SAMA in tuple format on OD Occlusion (0.013 vs. 0.004), OD Rotation (0.007 vs. 0.002), Test Set (0.010 vs. 0.002), and OD Size (0.008 vs. 0.003); Paligemma achieves higher tuple SAMA on five of the six columns. In Table 3, Qwen-VL 2B has lower center RMSE in tuple format on all six columns (e.g., 81.604 vs. 90.314 on OD Composition), and in Table 6, Paligemma and Qwen-VL 2B have consistently higher F1 in tuple format. The claims should be restricted to the 7B-8B models and to specific metrics, and the main text should explicitly acknowledge the counter-examples.
  2. [Section 3.4 and Section 7 (Limitations)] The central sentence-vs-tuple comparison is potentially confounded by the evaluation pipeline. SAMA assumes that outputs are structured enough for regex extraction and Jonker-Volgenant edit-distance matching, and the Limitations state that 'Regex patterns perform poorly for unstructured predictions.' Tuple and sentence formats fail in different ways: a single malformed token in a tuple can make an entire entry unparseable, while a sentence output may be partially recoverable even when it deviates from the template. Since both formats are scored by the same regex/JV machinery, the reported SAMA/RMSE advantage may reflect parser robustness rather than visual understanding. A parser-invariant check is needed, such as evaluating both formats with an LLM-based extractor or restricting the comparison to outputs that are parseable under both regex schemas.
  3. [Section 3.2 and Section 4.1] All quantitative conclusions rest on single runs with no confidence intervals, and the five OD test sets contain only 200 samples each. Many of the differences used to support the format claim are very small (e.g., Table 2: Qwen-VL 2B SAMA 0.048 vs. 0.054 on OD Composition; Phi-V 0.027 vs. 0.024). Without repeated seeds, error bars, or a significance test, these differences are within the range of run-to-run noise. At minimum, the paper should report the number of seeds and standard deviations, or explicitly temper claims that are based on such small absolute differences.
  4. [Section 4.2, Table 5, and Table S2] The loss-scaling claim is weakened by a selection effect and by inconsistent results. The scales 1.5, 2, 2.5, and 3.5 are evaluated on the same OD test sets that are used to report the final results, so the reported 'best' scale is selected on the test data rather than on a held-out validation set. Moreover, the improvement is not consistent: in Table 5, SW 2.0 improves center RMSE on OD Composition (31.81 vs. 34.05 for sentence baseline) but worsens it on OD Size (82.39 vs. 80.66), and Table S2 shows that scaling generally reduces SAMA accuracy. The paper should use a held-out validation split for scale selection and should quantify the accuracy trade-off before claiming that scaling 'enhances numerical approximation capabilities.'
  5. [Section 5 and Section 6] The discussion and conclusion overstate the generality of the findings. Section 5 says 'larger models fine-tuned in the sentence format outperform those fine-tuned in the tuple format by a wider margin compared to smaller models,' but even for MiniCPM-V2.6, Table 6 shows that tuple format yields higher or equal F1 on most OD sets (e.g., 0.742 vs. 0.733 on OD Composition, 0.804 vs. 0.814 on OD Occlusion, 0.809 vs. 0.810 on OD Rotation). The conclusion's statement that 'sentence-based formats consistently enhancing both numerical and spatial tasks' is therefore not supported by the full set of reported tables.
minor comments (5)
  1. [Table 3] The row label 'Ph3i-V' appears to be a typo for 'Phi3-V'; it should be consistent with Table 6.
  2. [Table 5] The column header 'OD Test Set' is ambiguous: it is not one of the five OD sets listed in Table 1 and appears to denote the in-domain test set. It should be labelled 'Test Set' consistently with Tables 2 and 3.
  3. [Section 3.4] The description of SAMA would benefit from explicit details about the regex patterns used for each attribute and about how edit-distance cost is computed across attributes of different types (discrete vs. continuous). Without these details, the metric is hard to reproduce independently.
  4. [Appendix A.2.2 and Table S4] Table S4 shows that sentence format is not consistently better on center-coordinate RMSE for the plant dataset (e.g., 152.31 vs. 121.78 on the train split), and the appendix attributes the discrepancy to linear matching. This should be acknowledged in the main text when citing Table 4, since the main text currently highlights only the count RMSE advantage.
  5. [Figures S5-S8] The qualitative outputs for Qwen-2B, Qwen-7B, Phi-3.5, and Paligemma contain many garbled tokens (e.g., 'Inoendo|1|>', 'Tquare', 'notcluded'). These figures are difficult to interpret and should either be replaced by cleaner representative outputs or annotated to show what is being compared.

Circularity Check

1 steps flagged · score 2.0 of 10

No central circularity; one peripheral post-hoc selection effect in the loss-scaling claim.

  1. fitted input called prediction [Section 4.2 (loss scaling discussion); Table 5 and Supplementary Table S2]
    "Finally, we examine the effect of scaled Cross Entropy (CE) Loss for numeric tokens in the LLM tokenizer. Table 5 demonstrates that scaling the CE loss to values of 2 or 2.5 enhances numerical approximation capabilities, although slight trade-offs in accuracy are observed in Supplementary Table S2."

    The scale values 1.5, 2, 2.5, and 3.5 were all evaluated on the same out-of-domain test sets whose numbers are then reported as the headline result. Choosing '2 or 2.5' as the beneficial scale is a post-hoc selection from that test-set sweep rather than a pre-specified prediction. The claim is therefore partly a restatement of the hyperparameter search: for example, in the 'Test Set' column of Table 5, SW2 (82.39) and SW2.5 (82.37) center-RMSE values are both worse than the unscaled sentence format (80.66), so the 'enhances' conclusion depends on selecting favorable columns. This is a mild selection effect rather than an equation-level identity, and it does not bear on the central sentence-versus-tuple comparison, so it is weighted lightly.

full rationale

The paper is an empirical benchmark and fine-tuning study rather than a derivation chain, so most circularity patterns do not apply. The central claim that sentence-formatted outputs outperform tuple formats is supported by fine-tuning several external VLMs and evaluating them with an external scoring pipeline (SAMA and RMSE) on a synthetic benchmark and a real-world Plant Phenotyping dataset. SAMA's dependence on regex extraction and Jonker-Volgenant edit-distance matching is a measurement confound, and the Limitations explicitly admit 'Regex patterns perform poorly for unstructured predictions'; however, this is not circular because SAMA is applied symmetrically to both formats and is not used to define the training targets or the output-format finding. The loss-scaling result is the only step resembling a fitted-input-called-prediction pattern: Table S2 sweeps scale factors 1.5, 2, 2.5, and 3.5 on the same OD test sets, and the '2 or 2.5' conclusion is selected from those very numbers. That is a real but mild evidential weakness, not a structural circularity. There is no load-bearing self-citation: the fine-tuning settings are borrowed from external MiniCPM-V work [23], and no uniqueness theorem or ansatz is imported from the authors' own prior publications. Overall score 2 reflects the single peripheral selection effect; the central format comparison stands on independent experiments.

Assumptions & free parameters 1 free parameters · 4 assumptions · 0 invented entities

No mathematical derivation is claimed; the circularity burden is low. The main free knob is the numeric-token loss scale, which was tuned on the reported test sets. The benchmark and metric introduce evaluation assumptions rather than physical entities.

free parameters (1)
  • numeric token loss scale = 2.0 or 2.5 (chosen after evaluating 1.5, 2, 2.5, 3.5 on OD test sets)
    Scales 1.5, 2, 2.5, 3.5 were evaluated on the same OD test sets used to report the gains; Table 5 and Table S2. Selecting the best scale after inspecting test performance makes the reported improvement partly fitted, not a parameter-free prediction.
assumptions (4)
  • domain assumption Synthetic 2D shapes on a 224x224 canvas are a valid proxy for 'primitive visual measurement understanding' relevant to real-world spatial tasks.
    The whole benchmark argument rests on this; Section 3.2 and the Limitations admit constrained variations.
  • domain assumption Regex-based extraction and JV edit-distance matching score sentence and tuple formats without systematic bias.
    SAMA assumes outputs are 'properly structured'; Limitations state regex performs poorly on unstructured predictions (Sections 3.4 and 7).
  • standard math Jonker-Volgenant assignment with edit distance finds the correct matching between predicted and ground-truth objects.
    Used as cost function in SAMA; no proof that edit distance minimizes attribute errors.
  • domain assumption A single fine-tuning run per configuration yields stable enough estimates to compare formats.
    Section 4.1 lists one run per setting; 200-sample OD sets make small differences potentially noise.

how reviews work

0 comments
Cite this review

Pith. "Pith review of Exploring Primitive Visual Measurement Understanding and the Role of Output Format in Learning in Vision-Language Models." pith.science (2026). https://pith.science/paper/BXKY2WHS

@misc{pith2026250115144,
  author       = {Pith},
  title        = {Pith review of: Exploring Primitive Visual Measurement Understanding and the Role of Output Format in Learning in Vision-Language Models},
  year         = {2026},
  howpublished = {\url{https://pith.science/paper/BXKY2WHS}},
  note         = {Machine review of arXiv:2501.15144}
}
read the original abstract

This work investigates the capabilities of current vision-language models (VLMs) in visual understanding and attribute measurement of primitive shapes using a benchmark focused on controlled 2D shape configurations with variations in spatial positioning, occlusion, rotation, size, and shape attributes such as type, quadrant, center-coordinates, rotation, occlusion status, and color as shown in Figure 1 and supplementary Figures S3-S81. We fine-tune state-of-the-art VLMs (2B-8B parameters) using Low-Rank Adaptation (LoRA) and validate them on multiple out-of-domain (OD) scenarios from our proposed benchmark. Our findings reveal that coherent sentence-based outputs outperform tuple formats, particularly in OD scenarios with large domain gaps. Additionally, we demonstrate that scaling numeric tokens during loss computation enhances numerical approximation capabilities, further improving performance on spatial and measurement tasks. These results highlight the importance of output format design, loss scaling strategies, and robust generalization techniques in enhancing the training and fine-tuning of VLMs, particularly for tasks requiring precise spatial approximations and strong OD generalization.

Figures

Figures reproduced from arXiv: 2501.15144 by the authors.

Figure 1
Figure 1. Sentence and Tuple Output Formats used for Fine [PITH_FULL_IMAGE:figures/full_fig_p002_1.png] view at source ↗
Figure 2
Figure 2. Structured Attribute Matching Accuracy Calculation for [PITH_FULL_IMAGE:figures/full_fig_p005_2.png] view at source ↗

Discussion (0). Continue with ORCID to comment.

Reference graph

Works this paper leans on

28 extracted references · 18 canonical work pages

  1. [1]

    Abdin, S

    M. Abdin, S. A. Jacobs, A. A. Awan, et al. Phi-3 techni- cal report: A highly capable language model locally on your phone. arXiv preprint arXiv:2404.14219, 2024. 2, 4

    arXiv 2024

  2. [2]

    Alayrac, J

    J.-B. Alayrac, J. Donahue, P. Luc, A. Miech, I. Barr, Y . Has- son, K. Lenc, A. Mensch, K. Millican, M. Reynolds, et al. Flamingo: a visual language model for few-shot learn- ing. Advances in neural information processing systems , 35:23716–23736, 2022. 3

    work page 2022

  3. [3]

    J. Bai, S. Bai, S. Yang, S. Wang, S. Tan, P. Wang, J. Lin, C. Zhou, and J. Zhou. Qwen-vl: A versatile vision-language model for understanding, localization, text reading, and be- yond. arXiv preprint arXiv:2308.12966, 1(2):3, 2023. 2

    arXiv 2023

  4. [4]

    Beyer, A

    L. Beyer, A. Steiner, A. S. Pinto, A. Kolesnikov, X. Wang, D. Salz, M. Neumann, I. Alabdulmohsin, M. Tschannen, E. Bugliarello, et al. Paligemma: A versatile 3b vlm for transfer. arXiv preprint arXiv:2407.07726, 2024. 2, 4

    arXiv 2024

  5. [5]

    Carion, F

    N. Carion, F. Massa, G. Synnaeve, N. Usunier, A. Kirillov, and S. Zagoruyko. End-to-end object detection with trans- formers. In European conference on computer vision, pages 213–229. Springer, 2020. 5

    work page 2020

  6. [6]

    D. F. Crouse. On implementing 2d rectangular assignment algorithms. IEEE Transactions on Aerospace and Electronic Systems, 52(4):1679–1696, 2016. 5

    work page 2016

  7. [7]

    Dubey, A

    A. Dubey, A. Jauhri, A. Pandey, A. Kadian, A. Al-Dahle, et al. The llama 3 herd of models. arXiv preprint arXiv:2407.21783, 2024. 3, 8

    arXiv 2024

  8. [8]

    D. A. Hudson and C. D. Manning. Gqa: A new dataset for real-world visual reasoning and compositional question answering. In Proceedings of the IEEE/CVF conference on computer vision and pattern recognition, pages 6700–6709,

    work page

Show all 28 references
  1. [9]

    H. Hyyr ¨o. Explaining and extending the bit-parallel approx- imate string matching algorithm of myers. Technical report, Citeseer, 2001. 5

    2001

  2. [10]

    Kv and A

    G. Kv and A. Mittal. Reducing language biases in vi- sual question answering with visually-grounded question en- coder. In Computer Vision–ECCV 2020: 16th European Conference, Glasgow, UK, August 23–28, 2020, Proceed- ings, Part XIII 16, pages 18–34. Springer, 2020. 3

    2020

  3. [11]

    H. Liu, C. Li, Q. Wu, et al. Visual instruction tuning. Ad- vances in neural information processing systems , 36, 2024. 2

    2024

  4. [12]

    Minervini, A

    M. Minervini, A. Fischbach, H. Scharr, et al. Finely-grained annotated datasets for image-based plant phenotyping. Pat- tern recognition letters, 81:80–89, 2016. 2, 4, 6, 7

    2016

  5. [13]

    Y . Qiao, H. Duan, X. Fang, J. Yang, L. Chen, S. Zhang, J. Wang, D. Lin, and K. Chen. Prism: A framework for decoupling and assessing the capabilities of vlms. arXiv preprint arXiv:2406.14544, 2024. 2

    2024 arXiv

  6. [14]

    Radford, J

    A. Radford, J. Wu, R. Child, D. Luan, D. Amodei, I. Sutskever, et al. Language models are unsupervised multi- task learners. OpenAI blog, 1(8):9, 2019. 2

    2019

  7. [15]

    Rahmanzadehgervi, L

    P. Rahmanzadehgervi, L. Bolton, M. R. Taesiri, and A. T. Nguyen. Vision language models are blind. arXiv preprint arXiv:2407.06581, 2024. 3

    2024 arXiv

  8. [16]

    R. Rivest. Rfc1321: The md5 message-digest algorithm,

  9. [17]

    Salewski, A

    L. Salewski, A. S. Koepke, H. P. Lensch, and Z. Akata. Clevr-x: A visual reasoning dataset for natural language ex- planations. In International Workshop on Extending Explain- able AI Beyond Deep Models and Classifiers , pages 69–88. Springer, 2020. 3

    2020

  10. [18]

    B. Shi, Z. Wu, M. Mao, X. Wang, and T. Darrell. When do we not need larger vision models?, 2024. 3

    2024

  11. [19]

    Singh, P

    S. Singh, P. Chaurasia, Y . Varun, P. Pandya, V . Gupta, V . Gupta, and D. Roth. Flowvqa: Mapping multimodal logic in visual question answering with flowcharts. arXiv preprint arXiv:2406.19237, 2024. 3

    2024 arXiv

  12. [20]

    Sinha, V

    N. Sinha, V . Jain, and A. Chadha. Guiding vision- language model selection for visual question-answering across tasks, domains, and knowledge types. arXiv preprint arXiv:2409.09269, 2024. 3

    2024 arXiv

  13. [21]

    P. Wang, S. Bai, S. Tan, S. Wang, Z. Fan, et al. Qwen2-vl: Enhancing vision-language model’s perception of the world at any resolution. arXiv preprint arXiv:2409.12191, 2024. 4

    2024 arXiv

  14. [22]

    A. Yang, B. Yang, B. Hui, B. Zheng, B. Yu, C. Zhou, C. Li, C. Li, D. Liu, F. Huang, et al. Qwen2 technical report. arXiv preprint arXiv:2407.10671, 2024. 3

    2024 arXiv

  15. [23]

    Y . Yao, T. Yu, et al. Minicpm-v: A gpt-4v level mllm on your phone. arXiv preprint arXiv:2408.01800, 2024. 2, 3, 4, 6 A. Appendix A.1. Ablation Studies Table S1 shows that fine-tuning only the vision head or the language model (LLM) component of the VLM re- sults in a decreas...

    2024 arXiv

  16. [24]

    Specifically, the shape limit is increased to 5–6 shapes per image, and the occlusion limit is also raised to 5–6 shapes

    OD Compositional : This scenario relaxes the con- straints on the number of shapes and the limit on over- lapping shapes from the training set. Specifically, the shape limit is increased to 5–6 shapes per image, and the occlusion limit is also raised to 5–6 shapes. This config...

  17. [25]

    This setup tests the model’s ability to detect and attribute more shapes in configurations that were not present in the training data

    OD Spatial Awareness: In this scenario, the number of shapes per image is similarly increased to 5–6, in- troducing new spatial arrangements. This setup tests the model’s ability to detect and attribute more shapes in configurations that were not present in the training data. ...

  18. [26]

    This scenario assesses the model’s ability to accurately detect and attribute shapes under previously unseen levels of occlusion

    OD Occlusion : While maintaining the same basic configuration as the training set, this scenario relaxes the occlusion limit, allowing for 4–5 shapes to over- lap simultaneously. This scenario assesses the model’s ability to accurately detect and attribute shapes under previou...

  19. [27]

    The ability to generalize to these out-of-domain rotations Model OD Comp

    OD Rotation: This scenario introduces additional ro- tation angles (45 ◦ and 72◦) for the shapes, presenting a challenge for the models to identify shapes in new orientations and accurately predict their angles. The ability to generalize to these out-of-domain rotations Model ...

  20. [28]

    Sentence Format Tuple Format MiniCPM-V2.5 Figure S4. Sentence Vs Tuple Output Comparison for MiniCPM-V2.5 Models for Validation Dataset

    OD Size: This scenario scales all shapes by a factor of 2, resulting in shapes that are twice their original size. It challenges the models to accurately locate the shapes and predict their attributes and coordinates un- der this new size configuration. Examples are provided i...

Pith tools

Reviewed August 10, 2026 · model on record in the stance chip above.