Pith. sign in

REVIEW 4 major objections 6 minor 29 references

You can get competitive 3D question answering from ordinary 2D vision-language models by orchestrating them in four modular stages—no 3D training required.

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 ↗

T0 review · grok-4.5

2026-07-31 07:14 UTC pith:OJSR747U

load-bearing objection Clean training-free modular 3D-QA pipeline with real numbers; competitiveness partly rides on frontier backbones and lexical normalization, and the BEV geometry does less work than the abstract implies. the 4 major comments →

arxiv 2607.28442 v1 pith:OJSR747U submitted 2026-07-30 cs.CV

ViewMind3D: Modular View-Aware Inference for Training-Free 3D-QA

Ping-Kun Chiang , Kun-Ru Wu , Po-han Li , Sandeep Chinchali , Ufuk Topcu , Yu-Chee Tseng This is my paper
classification cs.CV
keywords 3D question answeringtraining-freemulti-view reasoningvision-language modelsbird's-eye-viewvisual groundingembodied AIScanQA
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.

Most 3D question-answering systems need expensive 3D annotations and task-specific training, which limits how well they transfer to new rooms and robots. This paper argues that you do not need that pipeline. Instead, take multi-view RGB images of a scene, let a language model pick the views that matter for the question, mark the mentioned objects on those images, stamp each camera’s position and facing direction onto a simple bird’s-eye map, and then run a short chain of specialized reasoning roles to write the answer. On the standard ScanQA and SQA3D indoor benchmarks the resulting system matches or beats several trained 3D-language models, including strong gains on spatially grounded “What” questions, while never updating a single model weight. A sympathetic reader cares because the same general-purpose models already used for ordinary vision and chat can, with the right scaffolding, support embodied spatial reasoning in real environments.

Core claim

Effective 3D spatial question answering can be obtained without any 3D-specific training or full scene reconstruction by decomposing the task into four training-free stages—question-driven multi-view selection, language-guided object grounding, bird’s-eye-view camera indicators, and role-based answer synthesis—and orchestrating ordinary pretrained LLMs and VLMs over that structured representation.

What carries the argument

The ViewMind3D pipeline: a structured intermediate representation that packages filtered views, phrase-level object overlays, and BEV viewpoint markers, then feeds them through specialized agents (interpreter, captioner, summarizer, answerer, normalizer) so cross-view geometry is reasoned about without fine-tuning.

Load-bearing premise

Sparse multi-view RGB frames plus simple bird’s-eye pose overlays give the untuned 2D models enough geometric signal to reason reliably about 3D layout and relations.

What would settle it

On the same ScanQA/SQA3D splits, disable or scramble the BEV viewpoint indicators and guided boxes while keeping every other stage fixed; if accuracy on spatially grounded question types (especially “What” and “Which”) does not drop relative to the full pipeline, the claim that those spatial cues drive the 3D reasoning fails.

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

If this is right

  • Robotic perception stacks can add open-ended 3D-QA by calling existing LLMs/VLMs with multi-view images and poses, without collecting 3D-QA fine-tuning data.
  • Complete mesh or dense point-cloud reconstruction is not a prerequisite for competitive indoor 3D-QA when views are selected and annotated on the fly.
  • Spatially grounded question types benefit most from explicit object overlays and shared BEV references, so those modules can be prioritized when compute is limited.
  • The same modular scaffold can swap in different general-purpose backbones without retraining the 3D pipeline.

Where Pith is reading between the lines

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

  • If the bottleneck is sparse ScanNet-style viewpoints, denser or actively chosen camera sets should widen the gap over single-image baselines without any change to the reasoning agents.
  • Adaptive bypass of object grounding for count/procedure questions, already hinted at in the ablations, suggests a lightweight question-type router could raise overall accuracy with almost no extra cost.
  • The same four-stage pattern may transfer to outdoor or egocentric video QA where full 3D reconstruction is equally impractical.

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 / 6 minor

Summary. ViewMind3D proposes a fully training-free, modular pipeline for 3D question answering from multi-view RGB observations. It decomposes 3D-QA into (1) question-driven relevance view selection, (2) language-guided open-vocabulary object grounding (Florence-2), (3) BEV viewpoint indicators built from camera poses over a global map derived from the scene point cloud, and (4) multi-agent role-based answer generation (interpreter, captioner, summarizer, answerer, normalizer). Evaluated zero-shot on ScanQA val and SQA3D test with GPT-4.1 and OpenAI o3, the method reports 73.41 CIDEr on ScanQA and 50.75% average accuracy on SQA3D, competitive with fine-tuned 3D-LLMs and stronger than simple multi-view GPT baselines. Component ablations (Tables III–V) attribute gains to guided detection, BEV indicators, role decomposition, and term-based output normalization.

Significance. If the central claim holds—that modular orchestration of frozen general-purpose VLMs/LLMs can deliver genuine multi-view 3D spatial reasoning without 3D-specific training—the work is practically significant for embodied AI and robotics, where annotation cost and domain shift limit specialized 3D-LLMs. Strengths include a clear four-stage decomposition, public-benchmark evaluation under a stated zero-shot protocol, and systematic ablations isolating Guided Detection, Viewpoint Indicator, role decomposition, and Term-Based Instructions. The design is backbone-agnostic in principle and emphasizes intermediate structured representations, which aids interpretability. The contribution is primarily systems/engineering rather than a new learning principle; its lasting value depends on showing that the geometric and modular pieces, not just frontier VLM strength and lexical normalization, drive the reported gains.

major comments (4)
  1. [Abstract, §I, §III.A, §III.E] Abstract and §I claim reasoning “without requiring complete 3D reconstruction,” yet §III.A–E still take the ScanNet point cloud P and global camera poses C_i ∈ SE(3) as inputs to build B_globe and B_ind. This is lighter than full mesh/NeRF reconstruction but is not reconstruction-free geometry. The claim should be narrowed (e.g., “no learned 3D encoder / no dense reconstruction for reasoning”) and the dependence on posed RGB-D / point-cloud maps stated up front, including implications for true robotic deployment where global maps may be incomplete.
  2. [Table I, §III.F.5–6, §V.C] Headline competitiveness with fine-tuned 3D-LLMs conflates three factors: (i) stronger closed backbones (o3 vs. Flamingo/BLIP2-scale 3D-LLM), (ii) Term-Based Instructions / Answer Normalizer that force canonical ScanQA vocabulary (§III.F.5–6, §V.C), and (iii) the modular pipeline. Table III shows TBI alone lifts CIDEr 58.05→63.48 under GPT-4.1; the 73.41 figure uses o3. Please report a fixed-backbone comparison (same VLM for all zero-shot baselines and ViewMind3D) both with and without TBI/normalizer, and mark which metrics are n-gram-sensitive vs. semantic (BERTScore/Cos.Sim.), so the modular 3D-reasoning contribution is isolated from lexical alignment and model scale.
  3. [§III.E, §V.A, Tables III–IV] The paper’s own Viewpoint Indicator ablation (§V.A, Tables III–IV) states that frames_25k sparsity “limits the impact of camera pose cues” and that View Captioner semantic descriptors already partially substitute for BEV. Gains are real but modest and type-specific (CIDEr 57.41→63.48; SQA3D “Which” 30.20→37.89). Given that the central narrative is 3D spatial reasoning via BEV, please add a stronger geometry control: (a) shuffle or drop pose markers while keeping images, and/or (b) replace B_globe with a pose-free layout prior, under a fixed backbone. If performance stays near the full system, the geometric story should be tempered; if it drops, the BEV module is better supported.
  4. [§IV.E, §III.C] §IV.E reports ~90K tokens and ~120s per query, with the 3D-QA module (~35K) and Guided Detection (~30K) as bottlenecks. For a robotics-oriented training-free method this cost is load-bearing. Please add wall-clock and token comparisons against the GPT-4.1 MultiView / TwoStage baselines under matched APIs, and clarify whether Relevance View Selection’s recall-first policy has an upper bound on |I*| in practice. Without this, “practical for real-world environments” remains under-supported relative to simpler multi-view prompting.
minor comments (6)
  1. [Abstract, Table II] Table II format column is dense (V/S/Q/A/L); a short legend in the caption would help. Also align abstract “50.8%” / “73.4 CIDEr” with table values 50.75 and 73.41 for consistency.
  2. [§V.A, §III.D] §V.A notes adaptive bypass of Guided Detection for “How” questions and mixed compass/clock prompts by question type; these inference rules should be stated in §III or as an algorithm box so the main results are fully reproducible from the method section.
  3. [§III.D.3, §IV.A] Florence-2 NMS τ=0.6 (§III.D.3) is a free parameter; briefly justify or sensitivity-check. Same for the choice of scannet frames_25k sampling density.
  4. [§II] Related work cites GPT4Scene and Chat-Scene; please clarify more sharply what is training-free end-to-end versus methods that still use 3D-trained object encoders, to avoid overstating uniqueness.
  5. [Fig. 1] Fig. 1 is helpful but the numbering (2.1–2.3 then jump to 3–5) is slightly inconsistent with the four-component narrative in the abstract; consider aligning labels.
  6. [Table I, §III.F, References] Typos / style: “V oteNet” spacing in Table I; “F .” spacing before 3D-QA Module; arXiv IDs and access dates in references are fine but ensure consistent venue formatting.

Circularity Check

1 steps flagged

No load-bearing circular derivation; only minor metric-aligned output normalization (TBI) that boosts n-gram scores by lexical construction.

specific steps
  1. fitted input called prediction [§III.F.5–6 Answer Normalizer / Term-Based Instructions; §V.C Ablation on Term-Based Instructions; Table III]
    "Using the Term-Based Instructions setting, our agents are constrained to use a canonical vocabulary for object categories, colors, and spatial descriptors from frequently occurring terms in the human-annotated answers in ScanQA. As shown in Table III, this setting improves n-gram-based metrics (CIDEr 58.05→63.48)... Term-Based Instructions function as an output normalization mechanism that aligns model predictions with benchmark evaluation protocols."

    Canonical terms are taken from the evaluation set’s own frequent GT answers, then forced into model outputs; n-gram metrics (CIDEr/BLEU) therefore rise partly by lexical construction rather than by new spatial reasoning. The paper acknowledges this as normalization, and the core accuracy claim on SQA3D plus semantic metrics remain independently measured, so the step is minor and non-load-bearing for the training-free 3D-QA thesis.

full rationale

ViewMind3D is an engineering systems paper, not a first-principles derivation. Its central claim—that modular orchestration of frozen general-purpose LLMs/VLMs (view selection, Florence-2 grounding, BEV pose overlays, role-decomposed answer synthesis) yields competitive zero-shot 3D-QA—is tested on external public benchmarks (ScanQA val, SQA3D test) with third-party backbones (GPT-4.1, o3, Florence-2). There are no fitted physical/statistical parameters re-presented as predictions, no self-citation uniqueness theorems, and no equations that reduce to their own inputs by definition. The sole near-circular element is Term-Based Instructions / Answer Normalizer, which deliberately constrain ScanQA outputs to frequent ground-truth vocabulary and thereby raise CIDEr/BLEU by construction relative to unconstrained generation; the paper itself labels this output normalization for evaluation consistency (§III.F.5–6, §V.C, Table III: CIDEr 58.05→63.48). That is acknowledged metric alignment, not a circular proof of 3D reasoning. SQA3D accuracy (EM-style) is unaffected by TBI. Overall circularity is negligible.

Axiom & Free-Parameter Ledger

3 free parameters · 3 axioms · 1 invented entities

The central claim rests on the empirical adequacy of frozen general-purpose VLMs plus a small set of engineering choices (NMS threshold, BEV encoding style, role prompts, optional term normalization). No new physical or mathematical entities are postulated; the free parameters are ordinary hyper-parameters of the inference pipeline.

free parameters (3)
  • NMS IoU threshold τ = 0.6
    Fixed at 0.6 for suppressing overlapping detections; chosen by hand and affects which boxes reach the captioner.
  • BEV direction encoding scheme = compass/clock hybrid
    Compass vs clock markers selected per question type on SQA3D; an adaptive hybrid is suggested after seeing per-type differences.
  • Term-Based Instructions vocabulary = dataset-derived term list
    Canonical object/color/spatial terms mined from ScanQA human answers to raise n-gram overlap; directly improves CIDEr/BLEU.
axioms (3)
  • domain assumption General-purpose 2D VLMs/LLMs (GPT-4.1, o3, Florence-2) already encode sufficient spatial and semantic knowledge to support 3D-QA when given multi-view images, boxes, and BEV pose cues.
    Stated as the motivating question in §I and required for every zero-shot claim; never proved, only measured.
  • domain assumption Sparse RGB frames from ScanNet frames_25k plus camera extrinsics are an adequate proxy for the underlying 3D scene without full reconstruction.
    Problem statement §III.A and experimental setup §IV.A; ablation §V.A notes sparsity limits the value of pose cues.
  • ad hoc to paper Role decomposition into interpreter/captioner/summarizer/answerer/normalizer improves reasoning stability over a single monolithic agent.
    Introduced in §III.F and validated only by the paper’s own single- vs multi-agent ablation.
invented entities (1)
  • ViewMind3D structured representation (filtered views + annotated images + BEV indicators) no independent evidence
    purpose: Intermediate bridge that lets frozen 2D models perform cross-view 3D reasoning without parameter updates.
    Defined in §III.B as the key organizational device; no independent existence outside the pipeline.

pith-pipeline@v1.2.0-daily-grok45 · 17726 in / 2926 out tokens · 47214 ms · 2026-07-31T07:14:17.389489+00:00 · methodology

0 comments
read the original abstract

Recent advances in large language models (LLMs) and vision-language models (VLMs) have enabled new possibilities for 3D question answering (3D-QA), a key capability for embodied AI and robotic perception. However, most existing methods rely on 3D-specific training or fine-tuning with costly annotations, limiting their scalability and real-world applicability. We present \textbf{ViewMind3D}, a fully training-free and modular framework for 3D spatial reasoning over multi-view observations of a scene without requiring complete 3D reconstruction. The framework decomposes the 3D-QA task into four interpretable components: (1) question-driven multi-view selection, (2) guided visual grounding with language-conditioned object cues, (3) spatial context encoding via a bird's-eye-view (BEV) viewpoint indicator, and (4) structured answer generation through role-based reasoning. This design enables structured, robust, and interpretable reasoning without requiring model tuning. Experimental results on ScanQA and SQA3D show that ViewMind3D achieves competitive performance compared to prior training-free and fine-tuned 3D-LLMs. In particular, our method improves performance on spatially grounded question types, such as ``What'' questions in SQA3D, while maintaining strong overall accuracy (50.8\%) and achieving 73.4 CIDEr on ScanQA. These results demonstrate that effective 3D reasoning can be achieved through modular orchestration of general-purpose LLMs and VLMs for robotic perception in real-world environments.

Figures

Figures reproduced from arXiv: 2607.28442 by Kun-Ru Wu, Ping-Kun Chiang, Po-han Li, Sandeep Chinchali, Ufuk Topcu, Yu-Chee Tseng.

Figure 1
Figure 1. Figure 1: Overview of the ViewMind3D framework for 3D-QA. Given a natural language query, multi-view images, and a 3D point cloud, the system [PITH_FULL_IMAGE:figures/full_fig_p003_1.png] 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

29 extracted references · 4 linked inside Pith

  1. [1]

    3D-LLM: Injecting the 3D World into Large Language Models,

    Y . Hong, H. Zhen, P. Chen, S. Zheng, Y . Du, Z. Chen, and C. Gan, “3D-LLM: Injecting the 3D World into Large Language Models,” in Advances in Neural Information Processing Systems, vol. 36, 2023

    2023

  2. [2]

    PointLLM: Empowering Large Language Models to Understand Point Clouds,

    R. Xu, X. Wang, T. Wang, Y . Chen, J. Pang, and D. Lin, “PointLLM: Empowering Large Language Models to Understand Point Clouds,” in European Conference on Computer Vision (ECCV), 2024

    2024

  3. [3]

    ShapeLLM: Universal 3D Object Understanding for Embodied Interaction,

    Z. Qi, R. Dong, S. Zhang, H. Geng, C. Han, Z. Geet al., “ShapeLLM: Universal 3D Object Understanding for Embodied Interaction,” in European Conference on Computer Vision (ECCV), 2024

    2024

  4. [4]

    GPT-4V(ision) System Card,

    OpenAI, “GPT-4V(ision) System Card,” 2023, accessed: 2025-08-14. [Online]. Available: https://openai.com/index/gpt-4v-system-card

    2023

  5. [5]

    ScanQA: 3D Question Answering for Spatial Scene Understanding,

    D. Azuma, T. Miyanishi, S. Kurita, and M. Kawanabe, “ScanQA: 3D Question Answering for Spatial Scene Understanding,” inIEEE/CVF Conference on Computer Vision and Pattern Recognition (CVPR), 2022

    2022

  6. [6]

    SQA3D: Situated Question Answering in 3D Scenes,

    X. Ma, S. Yong, Z. Zheng, Q. Li, Y . Liang, S.-C. Zhu, and S. Huang, “SQA3D: Situated Question Answering in 3D Scenes,” inInterna- tional Conference on Learning Representations (ICLR), 2023

    2023

  7. [7]

    Grounded 3D-LLM with Referent Tokens,

    Y . Chen, S. Yang, H. Huang, T. Wang, R. Xu, R. Lyuet al., “Grounded 3D-LLM with Referent Tokens,”arXiv preprint arXiv:2405.10370, 2024

    Pith/arXiv arXiv 2024

  8. [8]

    Scene-LLM: Extending Language Model for 3D Visual Reasoning,

    R. Fu, J. Liu, X. Chen, Y . Nie, and W. Xiong, “Scene-LLM: Extending Language Model for 3D Visual Reasoning,” inIEEE/CVF Winter Conference on Applications of Computer Vision (WACV), 2025

    2025

  9. [9]

    GPT4Scene: Understand 3D Scenes from Videos with Vision-Language Models,

    Z. Qi, Z. Zhang, Y . Fang, J. Wang, and H. Zhao, “GPT4Scene: Understand 3D Scenes from Videos with Vision-Language Models,” arXiv preprint arXiv:2501.01428, 2024

    Pith/arXiv arXiv 2024

  10. [10]

    Chat-scene: Bridging 3d scene and large language models with object identifiers,

    H. Huang, Y . Chen, Z. Wang, R. Huang, R. Xu, T. Wang, L. Liu, X. Cheng, Y . Zhao, J. Panget al., “Chat-scene: Bridging 3d scene and large language models with object identifiers,”Advances in Neural Information Processing Systems (NeurIPS), 2024

    2024

  11. [11]

    Chat-Scene++: Exploiting Context-Rich Object Identification for 3D LLM,

    H. Huang, Y . Chen, Z. Wang, J. Pang, and Z. Zhao, “Chat-Scene++: Exploiting Context-Rich Object Identification for 3D LLM,”arXiv preprint arXiv:2603.27507, 2026

    Pith/arXiv arXiv 2026

  12. [12]

    LLM-Grounder: Open-V ocabulary 3D Visual Grounding With Large Language Model as an Agent,

    J. Yang, X. Chen, S. Qian, N. Madaan, M. Iyengar, D. F. Fouhey, and J. Chai, “LLM-Grounder: Open-V ocabulary 3D Visual Grounding With Large Language Model as an Agent,” inIEEE International Conference on Robotics and Automation (ICRA), 2024

    2024

  13. [13]

    [Online]

    OpenAI, “GPT-4,” 2023, accessed: 2025-08-14. [Online]. Available: https://openai.com/index/gpt-4-research/

    2023

  14. [14]

    Learning Transferable Visual Models From Natural Lan- guage Supervision,

    A. Radford, J. W. Kim, C. Hallacy, A. Ramesh, G. Goh, S. Agarwal et al., “Learning Transferable Visual Models From Natural Lan- guage Supervision,” inInternational Conference on Machine Learning (ICML), 2021

    2021

  15. [15]

    BLIP: Bootstrapping Language-Image Pre-training for Unified Vision-Language Under- standing and Generation,

    J. Li, D. Li, C. Xiong, and S. C. H. Hoi, “BLIP: Bootstrapping Language-Image Pre-training for Unified Vision-Language Under- standing and Generation,” inInternational Conference on Machine Learning (ICML), 2022

    2022

  16. [16]

    BLIP-2: Bootstrapping Language-Image Pre-training with Frozen Image Encoders and Large Language Models,

    J. Li, D. Li, S. Savarese, and S. C. H. Hoi, “BLIP-2: Bootstrapping Language-Image Pre-training with Frozen Image Encoders and Large Language Models,” inInternational Conference on Machine Learning (ICML), 2023

    2023

  17. [17]

    Visual Instruction Tuning,

    H. Liu, C. Li, Q. Wu, and Y . J. Lee, “Visual Instruction Tuning,” inAdvances in Neural Information Processing Systems (NeurIPS), vol. 36, 2023

    2023

  18. [18]

    LL3DA: Visual Interactive Instruction Tuning for Omni-3D Understanding Reasoning and Planning,

    S. Chen, X. Chen, C. Zhang, M. Li, G. Yu, H. Feiet al., “LL3DA: Visual Interactive Instruction Tuning for Omni-3D Understanding Reasoning and Planning,” inIEEE/CVF Conference on Computer Vision and Pattern Recognition (CVPR), 2024

    2024

  19. [19]

    DSPNet: Dual-vision Scene Perception for Robust 3D Question Answering,

    J. Luo, Y . Liu, W. Chen, Z. Li, Y . Wang, G. Li, and L. Lin, “DSPNet: Dual-vision Scene Perception for Robust 3D Question Answering,” in IEEE/CVF Conf. on Computer Vision and Pattern Recognition, 2025

    2025

  20. [20]

    Advancing 3D Scene Understanding with MV-ScanQA Multi-View Reasoning Evaluation and TripAlign Pre-training Dataset,

    W. Mo, Q. Chen, Y . Peng, S. Huang, and Y . Liu, “Advancing 3D Scene Understanding with MV-ScanQA Multi-View Reasoning Evaluation and TripAlign Pre-training Dataset,”ACM International Conference on Multimedia, 2025

    2025

  21. [21]

    ScanNet: Richly-Annotated 3D Reconstructions of In- door Scenes,

    A. Dai, A. X. Chang, M. Savva, M. Halber, T. Funkhouser, and M. Nießner, “ScanNet: Richly-Annotated 3D Reconstructions of In- door Scenes,” inIEEE Conference on Computer Vision and Pattern Recognition (CVPR), 2017

    2017

  22. [22]

    Florence-2: Advancing a Unified Representation for a Variety of Vision Tasks,

    B. Xiao, H. Wu, W. Xu, X. Dai, H. Hu, Y . Luet al., “Florence-2: Advancing a Unified Representation for a Variety of Vision Tasks,” in IEEE/CVF Conference on Computer Vision and Pattern Recognition (CVPR), 2024

    2024

  23. [23]

    ScanRefer: 3D Object Localization in RGB-D Scans Using Natural Language,

    D. Z. Chen, A. X. Chang, and M. Nießner, “ScanRefer: 3D Object Localization in RGB-D Scans Using Natural Language,” inEuropean Conference on Computer Vision (ECCV), 2020

    2020

  24. [24]

    Flamingo: A Visual Language Model for Few-Shot Learning,

    J.-B. Alayrac, J. Donahue, P. Luc, A. Miech, I. Barr, Y . Hassonet al., “Flamingo: A Visual Language Model for Few-Shot Learning,” in Advances in Neural Information Processing Systems (NeurIPS), 2022

    2022

  25. [25]

    BLEU: A Method for Automatic Evaluation of Machine Translation,

    K. Papineni, S. Roukos, T. Ward, and W.-J. Zhu, “BLEU: A Method for Automatic Evaluation of Machine Translation,” inAnnual Meeting of the Association for Computational Linguistics (ACL), 2002

    2002

  26. [26]

    METEOR: An Automatic Metric for MT Evaluation with Improved Correlation with Human Judgments,

    S. Banerjee and A. Lavie, “METEOR: An Automatic Metric for MT Evaluation with Improved Correlation with Human Judgments,” in ACL Workshop on Intrinsic and Extrinsic Evaluation Measures for Machine Translation and/or Summarization, 2005

    2005

  27. [27]

    ROUGE: A Package for Automatic Evaluation of Sum- maries,

    C.-Y . Lin, “ROUGE: A Package for Automatic Evaluation of Sum- maries,” inACL Workshop on Text Summarization Branches Out, 2004

    2004

  28. [28]

    CIDEr: Consensus-Based Image Description Evaluation,

    R. Vedantam, C. L. Zitnick, and D. Parikh, “CIDEr: Consensus-Based Image Description Evaluation,” inIEEE Conference on Computer Vision and Pattern Recognition (CVPR), 2015

    2015

  29. [29]

    BERTScore: Evaluating Text Generation with BERT,

    T. Zhang, V . Kishore, F. Wu, K. Q. Weinberger, and Y . Artzi, “BERTScore: Evaluating Text Generation with BERT,”arXiv preprint arXiv:1904.09675, 2019

    Pith/arXiv arXiv 1904