Pith. sign in

REVIEW 3 major objections 5 minor 25 references

Expand VSR Benchmark for VLLM to Expertize in Spatial Rules

T0 review · 3 major / 5 minor · reviewed 2026-08-11 · deepseek-v4-flash

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

Pith's one-line read A diffusion-and-encoder expansion of the VSR benchmark lifts VLLM spatial reasoning from 52% to 79% accuracy.

desk verdict Real and useful VSR recipe underneath a headline 27% claim that doesn't survive contact with the paper's own tables. read the letter →

arxiv 2412.18224 v1 pith:IEHJM7B3 submitted 2024-12-24 cs.CV cs.AI

Peijin Xie , Lin Sun , Bingquan Liu , Dexin Wang , Xiangzheng Zhang , Chengjie Sun , Jiajia Zhang This is my paper · ORCID
classification cs.CVcs.AI
keywords visualspatialreasoningvisionlargelanguagemodelsdiffusion-baseddataaugmentationmulti-encoderfusioninstructiontuninganswerbiasVSRbenchmarkSDXLrepainting
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 current vision-language models fail at visual spatial reasoning because they over-rely on language instructions and under-use visual positions, and that this can be fixed by expanding both sides of the training pipeline. It expands the VSR benchmark's text with 50 question templates and its images by repainting with the SDXL diffusion model in three modes, then merges four visual encoders (CLIP, SigLIP, DINOv2, SAM) into one vision stack. The resulting model, VSRE, scores 79% on the VSR test set versus 52% for LLaVA1.5 13B, a gain of over 27 percentage points, while also improving on spatial subsets of MME, MMBench, and SEEDv2. If right, the result shows that controlled diffusion-based data expansion and multi-encoder fusion are effective levers for relational reasoning in VLLMs.

What carries the argument

The load-bearing machinery is a two-arm expansion pipeline plus a merged vision encoder. The text arm rewrites VSR triplets into 50 template-based QA formats (30 hand-written, 20 GPT-4o generated) to reduce prompt sensitivity; the image arm uses SDXL to repaint seed images in three modes—image-to-image, text-to-image, and inpainting with bounding-box masks—to multiply visual examples while keeping captions fixed. The vision arm concatenates features from CLIP, SigLIP, DINOv2, and SAM after per-encoder projectors/adapters, so the LLM receives both language-aligned and self-supervised/segmentation-aligned visual tokens. These three arms together produce the VSRE model and its two test protocols, Test-G (random template) and Test-S (fixed template).

What would settle it

Have annotators verify a random sample of the generated pre-training images for whether the stated relation holds; if a large share are relation-inconsistent, retrain VSRE on only the verified subset and compare VSR accuracy. A drop would confirm the generated images are load-bearing; an unchanged score would show the gain comes from the templates or encoder fusion instead.

Watch

Extended reading notes

Core claim

On the paper's own terms, the central discovery is that a VLLM can be turned into a spatial-reasoning expert by expanding the VSR benchmark along two axes: rewriting every [subject, relation, object] triplet into dozens of instruction-answer pairs, and repainting the seed images with a diffusion model (image-to-image, text-to-image, and bounding-box inpainting) so that the model sees up to 100x more visual examples of the same relation. The second half of the recipe is architectural: freezing CLIP and adding SigLIP, DINOv2, and SAM features, each aligned by a projector/adapter and concatenated along the feature dimension, gives the model finer visual detail. The paper reports that VSRE reaches 79% accuracy on the VSR test set versus 52% for the LLaVA1.5 13B baseline, that t-SNE plots of its visual tokens separate spatial-relation categories more cleanly, and that the yes/no answer bias shrinks while scores on MME, MMBench, and SEEDv2 spatial subsets also rise.

Load-bearing premise

The whole gain rests on the assumption that diffusion-repainted images still contain the spatial relation named in the frozen caption, yet the paper applies no check—human or automatic—that the relation actually survives generation.

Editorial extensions

If this is right

  • VSRE's gains transfer to spatial subsets of MME, MMBench, and SEEDv2, suggesting the method is not overfit to VSR's formatting.
  • The open-sourced data and model let other teams apply the same diffusion-template-encoder recipe to other relation types or benchmarks.
  • The reduced yes/no bias indicates that visual attention shifts from entity co-occurrence to actual positional geometry, a claim that can be probed with attention maps.
  • Scaling pre-training data yields diminishing returns beyond roughly 400k images, so the practical recipe is pre-400k plus 500k instruction samples.

Reading between the lines

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

  • If the diffusion repainting does not reliably preserve the labeled relation, part of the 27-point gain could come from learning the template distribution rather than spatial geometry; a human-verified subset would isolate the two.
  • The same template-expansion trick could be applied to other VSR-style benchmarks (e.g., relationship questions in VQA) to diagnose prompt sensitivity before investing in image generation.
  • Because the four encoders are frozen, the architecture is cheap to reproduce; swapping in a depth or optical-flow encoder might give further spatial detail at the cost of inference speed.
  • The t-SNE clustering result is suggestive but not a metric of comprehension; a stricter test would ask VSRE to judge adversarial triplets where the same object pair appears in conflicting relations.
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 addresses visual spatial reasoning (VSR) for vision large language models (VLLMs). The authors introduce unified instruction test sets (Test-G and Test-S) derived from the VSR dataset, expand the training data through template-based text rewriting and diffusion-based image repainting, and merge four vision encoders (CLIP, SigLIP, DINOv2, SAM) into a single VLLM architecture. The resulting model, VSRE, is reported to achieve large gains on VSR and related subsets of MME, MMBench, and SEED, with a headline claim of a 27% accuracy increase over LLaVA1.5 13B on the VSR test set.

Significance. The paper's strengths are its concrete contributions: a reproducible protocol for expanding both text and image data for spatial reasoning, a systematic comparison of vision encoder combinations, and an open-sourced model and data release. If the reported gains survive a clean evaluation, the work would be a useful step toward making VLLMs more robust to instruction phrasing and more sensitive to spatial layout. The multi-encoder study is informative and the scaling experiments are extensive. However, the central quantitative claim is compromised by an incomparable baseline and by potential overlap between training and test templates, so the significance is not yet established.

major comments (3)
  1. [Conclusion; Table 1; Figures 5-6] The headline claim that 'VSRE surpasses the performance of LLaVA1.5 13B by 27% (from 52% to 79%) on VSR test set' is not supported by the reported numbers. The 52% figure appears in the captions of Figures 5 and 6 and refers to a 200-sample subset of 7 spatial relations, while the 79% figure is the Test-S accuracy of the four-encoder VSRE from Figure 4. Table 1 reports the zero-shot LLaVA1.5 13B on the full Test-S as 68.4 and on Test-G as 57.7; the improvement from the best data-only model to the four-encoder model is from 75.7 to 79 on Test-S and from 70.3 to 74.4 on Test-G. Please replace the 52%-to-79% comparison with an apples-to-apples comparison on the same test set, and adjust the abstract and conclusion accordingly.
  2. [Datasets; Training and Inference; Figure 4] The evaluation protocol has a circularity problem for the instruction-generalization claim. Test-G randomly samples a template from the same 50-template pool that is used to construct the turn-g 500k training data, and Test-S freezes the template '[caption], True or false,' which is also used in the turn-s 11k training set. In addition, the 'best-performing model' (VSRE) is selected by its Test-G/Test-S accuracy in Figure 4. Therefore, the claim that VSRE 'generalizes better to different instructions' has not been tested on held-out instruction formats. A clean test would reserve a disjoint set of templates (or template families) for evaluation, or the claim should be explicitly qualified as performance on instruction formats seen during training.
  3. [Expansion on Image Data] The diffusion-based image expansion lacks a quality check on the key property it is meant to enforce. The paper states that 'we freeze the text captions as prompts' and uses SDXL with image-to-image, text-to-image, and inpainting with bounding-box masks, but it does not report any human or automatic verification that the generated images still contain the labeled spatial relation. If a nontrivial fraction of generated images do not display the relation (e.g., 'in' rendered as 'next to'), then training on them may inject label noise rather than a stronger spatial prior, which would undermine the interpretation of the scaling gains. Please provide a verification protocol, such as human evaluation of a random sample or agreement with an existing VSR classifier, and report the consistency rate.
minor comments (5)
  1. [Table 1] The last row uses numeric subscripts (1,2,3) to denote training order, but the order is only explained in the text; please add an explicit notation explanation in the table caption.
  2. [Figure 4] The legend refers to 'rainbows' with dashed and solid lines, but without color it is difficult to distinguish the individual encoder combinations; adding line markers or a table of final accuracies would improve readability.
  3. [Table 3] Please report the number of samples in the 'relevant subsets' of MME, MMBench, and SEEDv2, and define which categories were selected, since the subset sizes are essential for interpreting the score differences.
  4. [Experiment] There are typos in the experimental section: 'tunning' should be 'tuning' and 'pertaining' should be 'pre-training' (e.g., in the 'Experiment' paragraph and the 'Expansion on Vision encoder' section).
  5. [Abstract] The claim of expanding spatially positioned image data with diffusion models 'for the first time' should either be supported by a literature search or softened, since the related-work section does not fully justify this novelty statement.

Circularity Check

2 steps flagged · score 4.0 of 10

The VSR-specific 27% claim is partly a test-selected maximum with a mismatched baseline, and template-pool reuse weakens the 'generalization' claim; external benchmarks keep the method from being fully circular.

  1. fitted input called prediction [Experiment: Scaling on Model (Figure 4); Conclusion]
    "Finally, we designated the best-performing model with 4 backbones trained by the total data obtained as a Visual Spatial Reasoning Expert(VSRE). ... Then, our proposed spatial expert VSRE surpasses the performance of LLaV A1.5 13B by 27% (from 52% to 79%) on VSR test set."

    VSRE is defined in the paper as the 'best-performing model' among the configurations compared in Figure 4, so the 79% Test-S number quoted in the abstract and conclusion is the maximum of a model-selection grid evaluated on the test set, not a prediction for a fixed architecture. The 52% baseline is taken from a different 200-sample, 7-relation subset (Figures 5-6), whereas the zero-shot LLaVA1.5 13B Test-S accuracy in Table 1 is 68.4. The 27% gain therefore mixes a test-selected maximum with a non-comparable subset baseline; the claimed improvement is not a statistically independent estimate.

  2. self definitional [Expansion for Spatial Expert: Expansion on Text Data; Experiment: Datasets]
    "Similarly, we expand the training set with the same template pool. ... Test-G random sampled a prompt from the 50 templates pool for each triplet to evaluate instruction-following generalization ability during spatial reasoning. ... IFT data: We used general 50 prompt templates (30 manual and 20 GPT4-generated) to expand the 11k triplet data nearly 50 times to 500k, then name it as 'turn-g 500k'. ... 'turn-s 11k' used the specific template same as Test-S."

    Test-G, which is used to support the claim of generalizing 'regardless of question style', is built by sampling from the same 50-template pool used to construct the turn-g 500k IFT training data. The test therefore draws instruction formats from the training distribution; it measures interpolation over templates, not generalization to unseen formats. In addition, turn-s 11k is trained with the exact same template as Test-S, so the Test-S improvement (75.7 to 76.6 in Table 1, and the 79% VSRE value) is partly a same-template training effect rather than improved spatial-relation perception.

full rationale

The paper's central VSR-specific headline is weakened by the two issues above, but the overall method is not circular in the strongest sense. The final VSRE is also evaluated on MME, MMBench and SEEDv2, which are external benchmarks with pre-existing questions; VSRE improves over LLaVA1.5 13B on these (155.00 vs 133.33 on MME, 64.8 vs 57.6 on MMBench, 46.6 vs 38.5 on SEEDv2). Those gains cannot be explained by model selection on the VSR test set or by template-pool reuse. The citations to prior multi-encoder work (MoF, MG-LLaVA, Cambrian, Prism) are external, not self-citations, and the VSR dataset itself is an external benchmark. The diffusion-generated image consistency issue is a real validity risk but is not a circularity: the labels are taken from the original VSR triplets, and the claim does not assume what it tries to prove. Overall, there is partial circularity and selection bias in the headline VSR comparison, but independent support for the central method remains, so the score is 4 rather than 6 or higher.

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

The paper introduces no new physical or conceptual entities. Its central claim rests instead on several empirical assumptions, most notably that diffusion-generated images preserve spatial labels and that test-set-driven model selection gives generalizable accuracy estimates.

free parameters (4)
  • Template pool size = 50 (30 manual + 20 GPT-4o)
    The 50 templates define Test-G and the IFT data; the split was chosen by the authors after their 69-template prompt experiments, not derived from theory.
  • Image augmentation ratio = 5:3:2 for image-to-image : text-to-image : inpainting
    Chosen by the authors to sequentially increase diversity; no ablations are reported for this ratio.
  • Pre-training data budget = pre-100k, pre-200k, pre-300k, pre-400k, pre-500k
    The best data scale (around 400k) is selected by observing Test-G and Test-S accuracy in Table 1 and Figure 4, making it fit to the evaluation set.
  • Vision encoder set = CLIP + SigLIP + DINOv2 + SAM (4 backbones)
    The four-encoder combination was chosen after ablation on Test-G and Test-S; incremental gains are small and the order of adding encoders was also chosen on test performance.
assumptions (5)
  • domain assumption The original VSR labels and train/test split are correct and unbiased.
    The paper builds all training and test data on the VSR triplets from Liu et al. without re-validating labels (Section "Datasets").
  • domain assumption SDXL-generated images preserve the intended spatial relation.
    The paper uses diffusion repainting with frozen captions but performs no filtering or human verification of generated images (Section "Expansion on Image Data").
  • domain assumption Forcing one-token answers and treating unlisted words as incorrect is a valid evaluation protocol.
    Inference is limited to one new word and any other response is judged incorrect (Section "Training and Inference"), which may undercount correct answers phrased differently.
  • domain assumption Test-G and Test-S accuracy is an appropriate model-selection criterion.
    Data and architecture choices are made from Table 1 and Figure 4, which report Test-G and Test-S; no separate validation split is mentioned.
  • domain assumption t-SNE visualizations and normalized intra-class distances reflect genuine spatial-relation understanding.
    The sensitivity analysis relies on clustering of visual tokens over 200 samples per relation (Section "More Sensitive Vision Features"), which is only indirect evidence.

how reviews work

0 comments
Cite this review

Pith. "Pith review of Expand VSR Benchmark for VLLM to Expertize in Spatial Rules." pith.science (2026). https://pith.science/paper/IEHJM7B3

@misc{pith2026241218224,
  author       = {Pith},
  title        = {Pith review of: Expand VSR Benchmark for VLLM to Expertize in Spatial Rules},
  year         = {2026},
  howpublished = {\url{https://pith.science/paper/IEHJM7B3}},
  note         = {Machine review of arXiv:2412.18224}
}
read the original abstract

Distinguishing spatial relations is a basic part of human cognition which requires fine-grained perception on cross-instance. Although benchmarks like MME, MMBench and SEED comprehensively have evaluated various capabilities which already include visual spatial reasoning(VSR). There is still a lack of sufficient quantity and quality evaluation and optimization datasets for Vision Large Language Models(VLLMs) specifically targeting visual positional reasoning. To handle this, we first diagnosed current VLLMs with the VSR dataset and proposed a unified test set. We found current VLLMs to exhibit a contradiction of over-sensitivity to language instructions and under-sensitivity to visual positional information. By expanding the original benchmark from two aspects of tunning data and model structure, we mitigated this phenomenon. To our knowledge, we expanded spatially positioned image data controllably using diffusion models for the first time and integrated original visual encoding(CLIP) with other 3 powerful visual encoders(SigLIP, SAM and DINO). After conducting combination experiments on scaling data and models, we obtained a VLLM VSR Expert(VSRE) that not only generalizes better to different instructions but also accurately distinguishes differences in visual positional information. VSRE achieved over a 27\% increase in accuracy on the VSR test set. It becomes a performant VLLM on the position reasoning of both the VSR dataset and relevant subsets of other evaluation benchmarks. We open-sourced the expanded model with data and Appendix at \url{https://github.com/peijin360/vsre} and hope it will accelerate advancements in VLLM on VSR learning.

Figures

Figures reproduced from arXiv: 2412.18224 by the authors.

Figure 1
Figure 1. The overall expansion method through the training process. On the text branch, questions and answers are rewritten by [PITH_FULL_IMAGE:figures/full_fig_p003_1.png] view at source ↗
Figure 2
Figure 2. Examples of 3 settings of image-to-image(first [PITH_FULL_IMAGE:figures/full_fig_p004_2.png] view at source ↗
Figure 4
Figure 4. Result of scaling vision model. We post the accu [PITH_FULL_IMAGE:figures/full_fig_p006_4.png] view at source ↗
Figures from the paper (1 more)
Figure 5
Figure 5. Figure 5: Distribution of selected 200 samples across 7 com [PITH_FULL_IMAGE:figures/full_fig_p007_5.png]

Discussion (0). Continue with ORCID to comment.

Reference graph

Works this paper leans on

25 extracted references · 1 canonical work pages

  1. [1]

    Chen, J.; Zhu, D.; Shen, X.; Li, X.; Liu, Z.; Zhang, P.; Krishnamoorthi, R.; Chandra, V.; Xiong, Y.; and Elhoseiny, M. 2023. MiniGPT-v2: large language model as a unified interface for vision-language multi-task learning. arXiv:2310.09478

    arXiv 2023

  2. [2]

    Dai, W.; Li, J.; Li, D.; Tiong, A. M. H.; Zhao, J.; Wang, W.; Li, B.; Fung, P.; and Hoi, S. 2023. InstructBLIP: Towards General-purpose Vision-Language Models with Instruction Tuning. arXiv:2305.06500

    arXiv 2023

  3. [3]

    Fu, C.; Chen, P.; Shen, Y.; Qin, Y.; Zhang, M.; Lin, X.; Yang, J.; Zheng, X.; Li, K.; Sun, X.; Wu, Y.; and Ji, R. 2024. MME: A Comprehensive Evaluation Benchmark for Multimodal Large Language Models. arXiv:2306.13394

    arXiv 2024

  4. [4]

    Goyal, Y.; Khot, T.; Summers-Stay, D.; Batra, D.; and Parikh, D. 2017. Making the v in vqa matter: Elevating the role of image understanding in visual question answering. In Proceedings of the IEEE conference on computer vision and pattern recognition, 6904--6913

    2017

  5. [5]

    Karamcheti, S.; Nair, S.; Balakrishna, A.; Liang, P.; Kollar, T.; and Sadigh, D. 2024. Prismatic vlms: Investigating the design space of visually-conditioned language models. arXiv preprint arXiv:2402.07865

    arXiv 2024

  6. [6]

    C.; Lo, W.-Y.; Dollár, P.; and Girshick, R

    Kirillov, A.; Mintun, E.; Ravi, N.; Mao, H.; Rolland, C.; Gustafson, L.; Xiao, T.; Whitehead, S.; Berg, A. C.; Lo, W.-Y.; Dollár, P.; and Girshick, R. 2023. Segment Anything. arXiv:2304.02643

    arXiv 2023

  7. [7]

    A.; et al

    Krishna, R.; Zhu, Y.; Groth, O.; Johnson, J.; Hata, K.; Kravitz, J.; Chen, S.; Kalantidis, Y.; Li, L.-J.; Shamma, D. A.; et al. 2017. Visual genome: Connecting language and vision using crowdsourced dense image annotations. International journal of computer vision, 123: 32--73

    2017

  8. [8]

    Li, B.; Wang, R.; Wang, G.; Ge, Y.; Ge, Y.; and Shan, Y. 2023. Seed-bench: Benchmarking multimodal llms with generative comprehension. arXiv preprint arXiv:2307.16125

    arXiv 2023

Show all 25 references
  1. [9]

    Lin, T.-Y.; Maire, M.; Belongie, S.; Hays, J.; Perona, P.; Ramanan, D.; Doll \'a r, P.; and Zitnick, C. L. 2014. Microsoft coco: Common objects in context. In Computer Vision--ECCV 2014: 13th European Conference, Zurich, Switzerland, September 6-12, 2014, Proceedings, Part V 1...

    2014

  2. [10]

    Liu, F.; Emerson, G.; and Collier, N. 2023. Visual Spatial Reasoning. arXiv:2205.00363

    2023 arXiv

  3. [11]

    Liu, H.; Li, C.; Li, Y.; and Lee, Y. J. 2024 a . Improved baselines with visual instruction tuning. In Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition, 26296--26306

    2024

  4. [12]

    Liu, H.; Li, C.; Li, Y.; and Lee, Y. J. 2024 b . Improved Baselines with Visual Instruction Tuning. arXiv:2310.03744

    2024 arXiv

  5. [13]

    Liu, H.; Li, C.; Wu, Q.; and Lee, Y. J. 2023 a . Visual Instruction Tuning. arXiv:2304.08485

    2023 arXiv

  6. [14]

    Liu, H.; Xue, W.; Chen, Y.; Chen, D.; Zhao, X.; Wang, K.; Hou, L.; Li, R.; and Peng, W. 2024 c . A survey on hallucination in large vision-language models. arXiv preprint arXiv:2402.00253

    2024 arXiv

  7. [15]

    Liu, Y.; Duan, H.; Zhang, Y.; Li, B.; Zhang, S.; Zhao, W.; Yuan, Y.; Wang, J.; He, C.; Liu, Z.; et al. 2023 b . Mmbench: Is your multi-modal model an all-around player? arXiv preprint arXiv:2307.06281

    2023 arXiv

  8. [16]

    McKinzie, B.; Gan, Z.; Fauconnier, J.-P.; Dodge, S.; Zhang, B.; Dufter, P.; Shah, D.; Du, X.; Peng, F.; Weers, F.; et al. 2024. Mm1: Methods, analysis & insights from multimodal llm pre-training. arXiv preprint arXiv:2403.09611

    2024 arXiv

  9. [17]

    Oquab, M.; Darcet, T.; Moutakanni, T.; Vo, H.; Szafraniec, M.; Khalidov, V.; Fernandez, P.; Haziza, D.; Massa, F.; El-Nouby, A.; Assran, M.; Ballas, N.; Galuba, W.; Howes, R.; Huang, P.-Y.; Li, S.-W.; Misra, I.; Rabbat, M.; Sharma, V.; Synnaeve, G.; Xu, H.; Jegou, H.; Mairal, ...

    2024 arXiv

  10. [18]

    Podell, D.; English, Z.; Lacey, K.; Blattmann, A.; Dockhorn, T.; Müller, J.; Penna, J.; and Rombach, R. 2023. SDXL: Improving Latent Diffusion Models for High-Resolution Image Synthesis. arXiv:2307.01952

    2023 arXiv

  11. [19]

    W.; Hallacy, C.; Ramesh, A.; Goh, G.; Agarwal, S.; Sastry, G.; Askell, A.; Mishkin, P.; Clark, J.; Krueger, G.; and Sutskever, I

    Radford, A.; Kim, J. W.; Hallacy, C.; Ramesh, A.; Goh, G.; Agarwal, S.; Sastry, G.; Askell, A.; Mishkin, P.; Clark, J.; Krueger, G.; and Sutskever, I. 2021. Learning Transferable Visual Models From Natural Language Supervision. arXiv:2103.00020

    2021 arXiv

  12. [20]

    C.; Yang, J.; Yang, S.; Iyer, A.; Pan, X.; et al

    Tong, S.; Brown, E.; Wu, P.; Woo, S.; Middepogu, M.; Akula, S. C.; Yang, J.; Yang, S.; Iyer, A.; Pan, X.; et al. 2024 a . Cambrian-1: A fully open, vision-centric exploration of multimodal llms. arXiv preprint arXiv:2406.16860

    2024 arXiv

  13. [21]

    Tong, S.; Liu, Z.; Zhai, Y.; Ma, Y.; LeCun, Y.; and Xie, S. 2024 b . Eyes wide shut? exploring the visual shortcomings of multimodal llms. In Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition, 9568--9578

    2024

  14. [22]

    Zhai, X.; Mustafa, B.; Kolesnikov, A.; and Beyer, L. 2023. Sigmoid Loss for Language Image Pre-Training. arXiv:2303.15343

    2023 arXiv

  15. [23]

    Zhao, X.; Li, X.; Duan, H.; Huang, H.; Li, Y.; Chen, K.; and Yang, H. 2024. MG-LLaVA: Towards Multi-Granularity Visual Instruction Tuning. arXiv preprint arXiv:2406.17770

    2024 arXiv

  16. [24]

    , " * write output.state after.block = add.period write newline

    ENTRY address archivePrefix author booktitle chapter edition editor eid eprint howpublished institution isbn journal key month note number organization pages publisher school series title type volume year label extra.label sort.label short.list INTEGERS output.state before.all...

  17. [25]

    write newline

    " write newline "" before.all 'output.state := FUNCTION n.dashify 't := "" t empty not t #1 #1 substring "-" = t #1 #2 substring "--" = not "--" * t #2 global.max substring 't := t #1 #1 substring "-" = "-" * t #2 global.max substring 't := while if t #1 #1 substring * t #2 gl...

Pith tools

Reviewed August 11, 2026 · model on record in the stance chip above.