Pith. sign in

REVIEW 4 major objections 4 minor 27 references

VT-LVLM-AR: A Video-Temporal Large Vision-Language Model Adapter for Fine-Grained Action Recognition in Long-Term Videos

T0 review · 4 major / 4 minor · reviewed 2026-08-05 · deepseek-v4-flash

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

Pith's one-line read A video-to-event token pipeline lets a frozen LLaVA-1.5 outperform dedicated action recognizers on NTU RGB+D and NTU RGB+D 120.

desk verdict Nice idea, missing interface: the paper never specifies how 1024-dim event tokens become LLaVA-1.5's visual input, so the frozen-core and SOTA claims don't yet hold up. read the letter →

arxiv 2508.15903 v1 pith:D3AKQBNJ submitted 2025-08-21 cs.CV

Kaining Li , Shuwei He , Zihan Xu This is my paper
classification cs.CV
keywords long-termvideoactionrecognitionlargevision-languagemodelsvisualeventtokensconceptualquantizationadaptivetemporalpoolingprompttuningNTURGB+Dparameter-efficientadaptation
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

VT-LVLM-AR's central claim is that a long video can be summarized as a short, language-like sequence of discrete "visual event tokens," and that a frozen Large Vision-Language Model can classify actions from that sequence as accurately as specialized models. The Video-to-Event Mapper learns these tokens through lightweight spatio-temporal feature extraction, adaptive temporal pooling, and vector quantization against a learned codebook, plus a coherence loss that keeps adjacent tokens narratively ordered. The frozen LLaVA-1.5 is then adapted with only 16 soft prompts via prompt tuning, leaving about 1.2 million trainable parameters. The paper reports state-of-the-art results on NTU RGB+D and NTU RGB+D 120, with the largest tested gains coming from the quantization step. A reader should care because the claim, if right, offers a parameter-efficient and more interpretable route from raw video into the reasoning machinery of a general vision-language model.

What carries the argument

The Video-to-Event Mapper (VTEM) is the load-bearing component. It converts raw frames into roughly 256 discrete event tokens by (1) extracting lightweight spatio-temporal segment features, (2) adaptively pooling those features over varying temporal windows, (3) quantizing each pooled feature to the nearest entry in a learned codebook of about 2048 visual concepts, and (4) applying an event-coherence bias through reconstruction and contrastive losses. The resulting visual event sentence is concatenated with a natural-language task instruction and 16 learnable soft-prompt tokens, and fed into a frozen LLaVA-1.5 model; only the soft prompts are updated.

What would settle it

Freeze every LLaVA-1.5 parameter, including the vision encoder and visual projector, train only the 16 soft prompts, and reproduce the 94.1% NTU RGB+D X-Sub result while logging the tensor path from the 1024-dimensional event tokens into the model. If a learned linear projection or image-patch embedding is required and updated, the frozen-core premise fails; if replacing LLaVA with a non-reasoning classifier on the same event tokens still gives about 94% accuracy, then the claimed contribution of LVLM reasoning is falsified.

Watch

Extended reading notes

Core claim

The paper claims that long videos can be reduced to a few hundred discrete tokens, called visual event sentences, and that reading those tokens with a frozen large vision-language model is enough to match or beat specialized action-recognition models. On NTU RGB+D X-Sub it reports 94.1%, on X-View 96.8%, and on NTU RGB+D 120 X-Sub 87.0% and X-Set 88.5%. The gain is attributed to the event mapping itself: without conceptual quantization accuracy falls from 94.1% to 91.5%, removing adaptive temporal pooling drops it to 92.8%, and dropping the event-coherence bias costs 0.8 points. The authors also report that prompt tuning matches full fine-tuning (94.1% vs 94.0%) while training only about 1.2

Load-bearing premise

The load-bearing premise is that the mapper's 256 discrete 1024-dimensional event tokens are directly accepted as visual input by the frozen LLaVA-1.5 model; the paper (Section III-B) never specifies the projection or interface, so if the tokens must pass through a learned projection or be embedded as image patches, the frozen-backbone claim collapses.

Editorial extensions

If this is right

  • If the result holds, action recognition no longer requires training a specialized video classification head; a frozen general-purpose vision-language model can classify actions from compact event tokens.
  • The same event-sentence interface could be reused for other video-language tasks, such as captioning or retrieval, by swapping the natural-language instruction.
  • The frozen-backbone plus prompt-tuning setup gives a practical parameter budget of about 1.2 million trainable parameters for adapting a multi-billion-parameter LVLM to video tasks.
  • The ablation identifies conceptual quantization as the most important ingredient, suggesting that the discrete event-token representation, rather than expensive full fine-tuning, drives the accuracy.
  • Performance peaks at 256 event tokens and a codebook of 2048; both over-compression and over-large vocabularies hurt, making event-sentence length and codebook size tunable hyperparameters.

Reading between the lines

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

  • Editorial inconsistency check: the claim of consistently state-of-the-art is not supported by the paper's own Table I on NTU RGB+D 120 X-Set, where PoseC3D is listed at 89.7% and VT-LVLM-AR at 88.5%; the text does not comment on this exception.
  • Editorial extension: the paper does not ablate the choice of LVLM itself; swapping LLaVA for a smaller LVLM or a non-reasoning classifier on the same tokens would reveal whether the gain comes from LLM-scale reasoning or from the tokenization.
  • Editorial extension: the reported margins over several baselines are small (0.4 to 0.8 points on some protocols), and the baseline set mixes skeleton and RGB methods; repeated-seed variance and a fine-grained long-video benchmark would strengthen the attribution.
  • Editorial extension: the coherence-loss ablation has the smallest effect in classification (0.8 points), so a more direct test of the visual event sentence idea would be to evaluate the same tokens on captioning or retrieval, where coherence should matter more.
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

4 major / 4 minor

Summary. VT-LVLM-AR proposes a two-stage action recognition pipeline: a Video-to-Event Mapper (VTEM) converts raw RGB video into M discrete 'visual event tokens' via a spatio-temporal extractor (left unspecified), adaptive temporal pooling, and codebook quantization; the resulting token sequence ('visual event sentence') is then fed to a frozen LLaVA-1.5 model adapted by P-Tuning v2 soft prompts for action classification. The paper reports 94.1% on NTU-60 X-Sub, 96.8% X-View, 87.0% on NTU-120 X-Sub, and 88.5% X-Set, together with ablations, a human evaluation of interpretability, and a parameter-efficiency analysis.

Significance. The paper targets an important and timely problem: adapting a powerful, frozen LVLM to long, fine-grained video action recognition through a compact intermediate representation. The modular idea of a discrete, temporally coherent 'visual event sentence' is appealing and, if carefully validated, could be a meaningful step toward interpretable and parameter-efficient video understanding. The manuscript has clear conceptual value, and the human evaluation of interpretability is a nice addition. However, as written, the central architecture is under-specified and the empirical claims are not well supported: there is no code, no error bars, and the baselines are predominantly skeleton-based. The strengths of the paper are therefore only potential; the current evidence is insufficient to verify the SOTA claim.

major comments (4)
  1. [III-B, Eqs. (3)-(5)] The core interface between VTEM and LLaVA-1.5 is not specified and appears architecturally incompatible. Section III-B says the visual event sentence E is 'directly fed as visual input' to LLaVA-1.5. LLaVA-1.5's visual pathway expects CLIP patch embeddings (e.g., 576 tokens from a ViT) and projects them with an MLP; it does not accept 256 arbitrary 1024-dimensional codebook vectors. If a trainable projection is inserted, the 'frozen LLaVA' claim and the 1.2M trainable-parameter count in Table IV are violated; if no projection is used, the described input cannot be ingested. The authors must specify the exact interface, including sequence-length handling and whether the vision tower is bypassed.
  2. [III-A, Eq. (4)] The VTEM training objective is not computable as stated. L_VTEM = αL_rec(P,E) + βL_cont(E) requires an explicit decoder that reconstructs pooled features P from event tokens E, but no decoder is described. In addition, the spatio-temporal feature extractor is left open ('can be based on architectures such as ...'), yet the experiments never name the actual extractor used. Without these details, the reported 94.1% result cannot be reproduced or independently verified.
  3. [IV-D, Tables II, V, VI] The ablation differences are very small (e.g., 94.1 vs. 94.0, 93.8, and 92.7) and no repeated runs, error bars, or statistical significance tests are reported. The conclusion that 'each component of the VTEM module is vital' is not supported by these deltas. At minimum, the authors should report mean±std over multiple seeds and provide significance tests. Also, choices of M and K are tuned on the validation set; the 'optimal' 94.1% should not be presented as evidence beyond the selected configuration.
  4. [IV-B, Table I] The state-of-the-art claim is not established by the comparison set. Table I lists ST-GCN, Shift-GCN, InfoGCN, PoseC3D, FR-Head, Koopman, GAP, HD-GCN, and STC-Net, which are predominantly skeleton-based. The paper does not compare against recent RGB-video or video-LVLM baselines. To support the claim of SOTA for video input, the authors need to include contemporary RGB/transformer and video-LLM methods under the same protocols.
minor comments (4)
  1. [References] References [8] and [9] do not point to LLaVA-1.5 or P-Tuning v2, respectively; the citations appear mismatched and should be corrected.
  2. [IV-F, Table IV] The reported 1.2M trainable parameters is not derived. With 16 soft prompt tokens inserted into specific layers, the count depends on the number of layers and hidden dimensions; please specify the calculation.
  3. [IV-E] The human evaluation lacks details: number of clips, annotator selection, blinding, and inter-annotator agreement are not reported, so the interpretability claim is difficult to assess.
  4. [Throughout] There are repeated typographical issues (e.g., 'LLaV A' instead of 'LLaVA') and inconsistent notation in Eqs. (1)-(2) (mixing h_j..h_k and p_i). A careful proofread is needed.

Circularity Check

0 steps flagged · score 0.0 of 10

No significant circularity: the framework is an empirical pipeline; reported gains come from standard training and model selection, not from a self-derived premise.

full rationale

I examined the derivation chain from the abstract through Sections III, IV, and V. The method is an empirical pipeline: raw video is mapped by VTEM (feature extraction, adaptive pooling, quantization, event-coherence losses) into visual event tokens, which are then passed to a frozen LLaVA-1.5 adapted with prompt tuning. The reported accuracies are measured on held-out NTU benchmarks, not derived from the model's own assumptions. The hyperparameters M (number of event tokens) and K (codebook size) are selected by validation-set sweeps (Tables V and VI) and then reported at their best values; this is standard model selection, not a fitted input being relabeled as a prediction. The ablation studies compare variants of the same model and do not assume the conclusion. No load-bearing self-citations appear: the cited LLaVA/P-Tuning references are external and, even if mis-cited, do not reduce the method's result to the citation. The skeptic's concern about the LLaVA-1.5 interface (Section III-B says the visual event sentence is 'directly fed' to LLaVA-1.5 while Section IV-A specifies 1024-dimensional tokens 'designed to match the visual input dimension expected by the subsequent LLaVA-1.5 model') is an architectural reproducibility or correctness issue: if no projection is specified, the frozen-core claim may be unverifiable, but this does not make any equation or result equivalent to its own input by construction. There is no self-definitional reduction, no renamed known result presented as unification, and no uniqueness theorem imported from the authors. The central claims are empirical and falsifiable against independent benchmarks, so the circularity score is 0.

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

The framework's success depends on several unverified premises: that quantized pooling features preserve action-discriminative details, that the frozen LLaVA can accept these tokens as visual input, and that the prompt tuning is sufficient. The hyperparameters M and K are tuned on the benchmark, so the reported peak accuracy is a fit to the data.

free parameters (4)
  • M (number of visual event tokens) = 256 (default; 64-512 swept in Table V)
    Sequence length of the visual event sentence; tuned on NTU X-Sub.
  • K (codebook size) = 2048 (optimal; 512-4096 swept in Table VI)
    Vocabulary size of conceptual quantization; tuned on NTU X-Sub.
  • L (soft prompt length) = 16 tokens
    Number of learnable soft prompts inserted into LLaVA; chosen in Section IV-A.
  • alpha, beta (loss weights) = not reported
    Hyperparameters in Eq. (4) balancing reconstruction and contrastive losses; values not given.
assumptions (4)
  • standard math Straight-through estimator enables gradient flow through the argmin in Eq. (3).
    Invoked in Section III-A without proof or ablation; standard for VQ but affects codebook learning.
  • domain assumption The NTU RGB+D and NTU-120 benchmarks are used with correct protocols and the reported baseline accuracies are accurate.
    The paper compares against 'best-reported accuracies' of prior methods but does not reproduce them.
  • domain assumption Frozen LLaVA-1.5 can ingest a sequence of 256 codebook vectors of dimension 1024 as its visual input.
    The paper does not describe the projection or interface; LLaVA normally consumes CLIP patch embeddings.
  • ad hoc to paper Adaptive temporal pooling and the event coherence bias produce tokens that correspond to meaningful sub-actions.
    Only supported by the paper's own human evaluation, which lacks statistical rigor.
invented entities (1)
  • Visual event token / visual event sentence
    purpose: Discrete representation bridging video frames and LLaVA input
    Introduced by the paper; interpretability claim rests on a small internal human study with no external validation.

how reviews work

0 comments
Cite this review

Pith. "Pith review of VT-LVLM-AR: A Video-Temporal Large Vision-Language Model Adapter for Fine-Grained Action Recognition in Long-Term Videos." pith.science (2026). https://pith.science/paper/D3AKQBNJ

@misc{pith2026250815903,
  author       = {Pith},
  title        = {Pith review of: VT-LVLM-AR: A Video-Temporal Large Vision-Language Model Adapter for Fine-Grained Action Recognition in Long-Term Videos},
  year         = {2026},
  howpublished = {\url{https://pith.science/paper/D3AKQBNJ}},
  note         = {Machine review of arXiv:2508.15903}
}
read the original abstract

Human action recognition in long-term videos, characterized by complex backgrounds and subtle action differences, poses significant challenges for traditional deep learning models due to computational overhead, difficulty in capturing long-range temporal dependencies, and limited semantic understanding. While Large Language Models (LLMs) and Large Vision-Language Models (LVLMs) have shown remarkable capabilities in multi-modal understanding and reasoning, their direct application to continuous video streams for fine-grained action recognition remains an open problem. This paper introduces VT-LVLM-AR (Video-Temporal Large Vision-Language Model Adapter for Action Recognition), a novel framework designed to bridge this gap. VT-LVLM-AR comprises a Video-to-Event Mapper (VTEM) that efficiently transforms raw video into compact, semantically rich, and temporally coherent "visual event sequences" through lightweight spatio-temporal feature extraction, adaptive temporal pooling, and conceptual quantization with an event coherence bias. These visual event sequences are then fed into an LVLM-based Action Reasoning module, specifically a frozen LLaVA-1.5 model, adapted using parameter-efficient Prompt Tuning (P-Tuning v2) for action classification. Comprehensive evaluations on the NTU RGB+D and NTU RGB+D 120 datasets demonstrate that VT-LVLM-AR consistently achieves state-of-the-art performance, surpassing existing methods (e.g., 94.1% accuracy on NTU RGB+D X-Sub). Ablation studies confirm the critical contributions of VTEM's components and the efficacy of Prompt Tuning, while human evaluations underscore the interpretability of our visual event representations. This work highlights the immense potential of leveraging LVLMs for robust and interpretable video action understanding through effective video-to-language translation and efficient model adaptation.

Discussion (0). Sign in to comment.

Reference graph

Works this paper leans on

27 extracted references · 19 canonical work pages

  1. [1]

    Human action recognition from various data modalities: A review,

    Z. Sun, Q. Ke, H. Rahmani, M. Bennamoun, G. Wang, and J. Liu, “Human action recognition from various data modalities: A review,” IEEE Trans. Pattern Anal. Mach. Intell. , pp. 3200–3225, 2023

    work page 2023

  2. [2]

    2d versus 3d convolutional spiking neural networks trained with unsupervised STDP for human ac- tion recognition,

    M. el Assal, P. Tirilly, and I. M. Bilasco, “2d versus 3d convolutional spiking neural networks trained with unsupervised STDP for human ac- tion recognition,” in International Joint Conference on Neural Networks, IJCNN 2022, Padua, Italy, July 18-23, 2022 . IEEE, 2022, pp. 1–8

    work page 2022

  3. [3]

    Overview of the transformer-based models for NLP tasks,

    A. Gillioz, J. Casas, E. Mugellini, and O. A. Khaled, “Overview of the transformer-based models for NLP tasks,” in Proceedings of the 2020 Federated Conference on Computer Science and Information Systems, FedCSIS 2020, Sofia, Bulgaria, September 6-9, 2020 , 2020, pp. 179– 183

    work page 2020

  4. [4]

    A survey on large language model (LLM) security and privacy: The good, the bad, and the ugly,

    Y . Yao, J. Duan, K. Xu, Y . Cai, E. Sun, and Y . Zhang, “A survey on large language model (LLM) security and privacy: The good, the bad, and the ugly,” CoRR, 2023

    2023

  5. [5]

    Lvlm-ehub: A comprehensive evaluation bench- mark for large vision-language models,

    P. Xu, W. Shao, K. Zhang, P. Gao, S. Liu, M. Lei, F. Meng, S. Huang, Y . Qiao, and P. Luo, “Lvlm-ehub: A comprehensive evaluation bench- mark for large vision-language models,” IEEE Trans. Pattern Anal. Mach. Intell., pp. 1877–1893, 2025

    work page 2025

  6. [6]

    Visual in-context learning for large vision-language models,

    Y . Zhou, X. Li, Q. Wang, and J. Shen, “Visual in-context learning for large vision-language models,” in Findings of the Association for Computational Linguistics, ACL 2024, Bangkok, Thailand and virtual meeting, August 11-16, 2024 . Association for Computational Linguis- tics, 2024, pp. 15 890–15 902

    2024

  7. [7]

    Weak to strong generalization for large language models with multi-capabilities,

    Y . Zhou, J. Shen, and Y . Cheng, “Weak to strong generalization for large language models with multi-capabilities,” in The Thirteenth International Conference on Learning Representations , 2025

    2025

  8. [8]

    Llava-more: A comparative study of llms and visual backbones for enhanced visual instruction tuning,

    F. Cocchi, N. Moratelli, D. Caffagni, S. Sarto, L. Baraldi, M. Cornia, and R. Cucchiara, “Llava-more: A comparative study of llms and visual backbones for enhanced visual instruction tuning,” CoRR, 2025

    work page 2025

Show all 27 references
  1. [9]

    Adaptive prompt: Unlocking the power of visual prompt tuning,

    M. Le, A. Nguyen, H. Nguyen, C. Nguyen, and N. Ho, “Adaptive prompt: Unlocking the power of visual prompt tuning,” CoRR, 2025

    2025

  2. [10]

    NTU RGB+D: A large scale dataset for 3d human activity analysis,

    A. Shahroudy, J. Liu, T. Ng, and G. Wang, “NTU RGB+D: A large scale dataset for 3d human activity analysis,” in 2016 IEEE Conference on Computer Vision and Pattern Recognition, CVPR 2016, Las Vegas, NV , USA, June 27-30, 2016. IEEE Computer Society, 2016, pp. 1010–1019

    2016

  3. [11]

    Human action recognition and prediction: A survey,

    Y . Kong and Y . Fu, “Human action recognition and prediction: A survey,” CoRR, 2018

    2018

  4. [12]

    A comprehensive study of deep video action recognition,

    Y . Zhu, X. Li, C. Liu, M. Zolfaghari, Y . Xiong, C. Wu, Z. Zhang, J. Tighe, R. Manmatha, and M. Li, “A comprehensive study of deep video action recognition,” CoRR, 2020. 8

    2020

  5. [13]

    Action recognition based on efficient deep feature learning in the spatio-temporal domain,

    F. Husain, B. Dellen, and C. Torras, “Action recognition based on efficient deep feature learning in the spatio-temporal domain,” IEEE Robotics Autom. Lett. , pp. 984–991, 2016

    2016

  6. [14]

    Cross-fiber spatial-temporal co-enhanced networks for video action recognition,

    H. Wu, Z.-J. Zha, X. Wen, Z. Chen, D. Liu, and X. Chen, “Cross-fiber spatial-temporal co-enhanced networks for video action recognition,” in Proceedings of the 27th ACM international conference on multimedia , 2019, pp. 620–628

    2019

  7. [15]

    Mutually reinforced spatio-temporal convolutional tube for human action recognition

    H. Wu, J. Liu, Z.-J. Zha, Z. Chen, and X. Sun, “Mutually reinforced spatio-temporal convolutional tube for human action recognition.” in IJCAI, 2019, pp. 968–974

    2019

  8. [16]

    Multi-scale spatial- temporal integration convolutional tube for human action recognition,

    H. Wu, J. Liu, X. Zhu, M. Wang, and Z.-J. Zha, “Multi-scale spatial- temporal integration convolutional tube for human action recognition,” in Proceedings of the Twenty-Ninth International Conference on Inter- national Joint Conferences on Artificial Intelligence, 2021, pp. 753–759

    2021

  9. [17]

    Long-term temporal convolutions for action recognition,

    G. Varol, I. Laptev, and C. Schmid, “Long-term temporal convolutions for action recognition,” IEEE Trans. Pattern Anal. Mach. Intell. , pp. 1510–1517, 2018

    2018

  10. [18]

    Finegym: A hierarchical video dataset for fine-grained action understanding,

    D. Shao, Y . Zhao, B. Dai, and D. Lin, “Finegym: A hierarchical video dataset for fine-grained action understanding,” in 2020 IEEE/CVF Conference on Computer Vision and Pattern Recognition, CVPR 2020, Seattle, WA, USA, June 13-19, 2020 . Computer Vision Foundation / IEEE, 2020...

    2020

  11. [19]

    End-to-end video-level representation learning for action recognition,

    J. Zhu, Z. Zhu, and W. Zou, “End-to-end video-level representation learning for action recognition,” in 24th International Conference on Pattern Recognition, ICPR 2018, Beijing, China, August 20-24, 2018 . IEEE Computer Society, 2018, pp. 645–650

    2018

  12. [20]

    Stnet: Local and global spatial-temporal modeling for action recogni- tion,

    D. He, Z. Zhou, C. Gan, F. Li, X. Liu, Y . Li, L. Wang, and S. Wen, “Stnet: Local and global spatial-temporal modeling for action recogni- tion,” in The Thirty-Third AAAI Conference on Artificial Intelligence, AAAI 2019, The Thirty-First Innovative Applications of Artificial I...

    2019

  13. [21]

    A survey on efficient vision-language models,

    G. Shinde, A. Ravi, E. Dey, S. Sakib, M. Rampure, and N. Roy, “A survey on efficient vision-language models,” WIREs Data. Mining. Knowl. Discov., 2025

    2025

  14. [22]

    Cheap and quick: Efficient vision-language instruction tuning for large language models,

    G. Luo, Y . Zhou, T. Ren, S. Chen, X. Sun, and R. Ji, “Cheap and quick: Efficient vision-language instruction tuning for large language models,” in Advances in Neural Information Processing Systems 36: Annual Conference on Neural Information Processing Systems 2023, NeurIPS 20...

    2023

  15. [23]

    PEFT A2Z: parameter-efficient fine-tuning survey for large language and vision models,

    N. J. Prottasha, U. R. Chowdhury, S. Mohanto, T. Nuzhat, A. A. Sami, M. S. Ali, M. S. I. Sobuj, H. Raman, M. Kowsher, and O. O. Garibay, “PEFT A2Z: parameter-efficient fine-tuning survey for large language and vision models,” CoRR, 2025

    2025

  16. [24]

    Visualgpt: Data- efficient adaptation of pretrained language models for image captioning,

    J. Chen, H. Guo, K. Yi, B. Li, and M. Elhoseiny, “Visualgpt: Data- efficient adaptation of pretrained language models for image captioning,” in IEEE/CVF Conference on Computer Vision and Pattern Recognition, CVPR 2022, New Orleans, LA, USA, June 18-24, 2022 . IEEE, 2022, pp. 1...

    2022

  17. [25]

    Multi-modal large language models are effective vision learners,

    L. Sun, C. Ahuja, P. Chen, M. D’Zmura, K. Batmanghelich, and P. Bontrager, “Multi-modal large language models are effective vision learners,” in IEEE/CVF Winter Conference on Applications of Computer Vision, WACV 2025, Tucson, AZ, USA, February 26 - March 6, 2025 . IEEE, 2025,...

    2025

  18. [26]

    Improving medical large vision- language models with abnormal-aware feedback,

    Y . Zhou, L. Song, and J. Shen, “Improving medical large vision- language models with abnormal-aware feedback,” arXiv preprint arXiv:2501.01377, 2025

    2025 arXiv

  19. [27]

    Aligngpt: Multi-modal large language models with adaptive alignment capability,

    F. Zhao, T. Pang, C. Li, Z. Wu, J. Guo, S. Xing, and X. Dai, “Aligngpt: Multi-modal large language models with adaptive alignment capability,” CoRR, 2024. 9

    2024

Pith tools

Reviewed August 5, 2026 · model on record in the stance chip above.