Pith. sign in

REVIEW 3 major objections 5 minor 136 references

This paper claims that drawing ground-truth boxes on the training image and distilling from a frozen teacher that sees that overlay improves visual grounding accuracy of coordinate-generating multimodal language models, while leaving infere

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 · deepseek-v4-flash

2026-08-01 22:28 UTC pith:OSFT67S7

load-bearing objection Sensible privileged-distillation recipe for grounding, but the key mechanism gains rest on single-run differences near noise level. the 3 major comments →

arxiv 2607.15732 v1 pith:OSFT67S7 submitted 2026-07-17 cs.CV

IoUPD: IoU-Aware Privileged Distillation for Visual Grounding with Multimodal Large Language Models

Xiuyuan Zhu , Ke Lu , Hao Wu , Zijin Du , Dongming Zhang , Jian Xue This is my paper
classification cs.CV
keywords visual groundingprivileged distillationIoU-aware weightingcoordinate generationmultimodal large language modelsknowledge distillationreferring expression comprehension
verification ladder T0 review T1 audit T2 compute T3 formal T4 reserved

The pith

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

The paper targets the mismatch between token-level training and box-overlap evaluation in multimodal language models that emit bounding boxes as text. It claims that using the ground-truth box as more than a label—drawing it on the training image for a frozen teacher and distilling that teacher's predictions into the student—raises region-level grounding accuracy by several absolute points on standard referring-expression benchmarks, with no change to the inference interface. A reader should care because this is a training-only intervention that improves localization without adding modules, heads, or deployment-time computation.

Core claim

On its own terms, IoU-PD establishes that ground-truth bounding boxes can serve as privileged training-time guidance beyond coordinate targets. During training the student sees the original image and prompt, while a frozen teacher sees the same image with the ground-truth box drawn in green and a prompt that appends 'The answer is located within the green rectangle.' The student is trained with a supervised fine-tuning loss on the ground-truth coordinate string plus an on-policy KL distillation loss over the student's own decoded tokens, weighted token-by-token by IoU-dependent, coordinate-error, digit-position, agreement, and confidence factors. At inference the teacher, overlay, and hint a

What carries the argument

The load-bearing mechanism is the privileged teacher input: the ground-truth box drawn as a green rectangle on the image plus a textual hint appended to the prompt. This is paired with a weighted KL distillation objective on the student's own decoded tokens, L = L_kd + alpha L_sft, where L_kd = sum_t w_t KL(stop_grad(teacher(·|x_t, prefix)) || student(·|x_s, prefix)). The token weight w_t = r(u) · c_t · d_t · a_t · e_t combines a sample-level exponential IoU factor, a softmax over coordinate errors, digit-position significance, a sigmoid teacher–student agreement gate, and an entropy-based teacher confidence term. The SFT anchor keeps the student tied to the ground-truth string; the weights

Load-bearing premise

The entire gain rests on the assumption that a frozen teacher conditioned on a box-marked image and hint produces a target distribution close enough to the student's own inference-time conditional distribution that distilling those targets with IoU-based weights improves the student.

What would settle it

Run the same distillation but feed the teacher the original image plus a semantically correct yet visually non-informative hint (e.g., 'the answer is in the central region') with no overlay. If the mIoU gain over SFT mostly disappears, the improvement is caused by the overlay rather than by distillation itself. More directly, on a held-out split, compute per-example KL divergence between the frozen teacher's distribution on box-marked inputs and the student's distribution on unmarked inputs; if examples with large mismatch still improve substantially, the proposed transferability mechanism is

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

If this is right

  • Any coordinate-generating multimodal model can be improved without changing its inference interface by reusing ground-truth boxes during fine-tuning.
  • No extra prediction head, EMA teacher, or separately trained teacher is needed; a frozen copy of the same model suffices for the teacher branch.
  • The gains grow under stricter IoU thresholds (P@0.95 improves by more than twelve points in the reported setting), meaning the intervention sharpens already-correct predictions rather than only fixing gross misses.
  • The benefit appears across object sizes, with largest absolute gains on small and medium objects, where token-level supervision is least informative.
  • Removing the SFT anchor sharply reduces the gain, implying privileged distillation complements rather than replaces direct coordinate supervision.

Where Pith is reading between the lines

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

  • The component ablations imply that the box overlay itself, not distillation alone, carries most of the improvement; a natural follow-up would replace the green rectangle with softened or semantic overlays to test whether the effect is the geometric region or the visual cue.
  • Because the teacher is frozen and sees the same scene plus a region marker, the recipe could transfer to other structured outputs such as keypoints, polygons, or dense captions, where token-position error similarly mismatches the evaluation metric.
  • The paper acknowledges the teacher–student conditional mismatch; a per-example diagnostic comparing teacher confidence against student improvement would clarify whether that mismatch caps the gains and could motivate a fidelity-gated weight.
  • The dependence on ground-truth boxes limits the method to supervised datasets; a pseudo-box or self-training variant would test whether the privileged guidance can be obtained without extra annotations.

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

3 major / 5 minor

Summary. The paper proposes IoU-PD, a training-time method for coordinate-generating multimodal large language models in visual grounding. A student receives the original image and referring expression, while a frozen teacher receives an image with the ground-truth box drawn in green and an augmented prompt containing the hint 'The answer is located within the green rectangle.' The total objective combines an SFT anchor on the ground-truth coordinate string with a token-weighted KL distillation loss on student-decoded trajectories. The token weights incorporate sample-level IoU, coordinate error, digit position, teacher–student agreement, and teacher confidence. At inference the teacher branch and privileged inputs are removed, so the inference interface is unchanged. Experiments on RefCOCO/RefCOCO+/RefCOCOg report gains over the Qwen3-VL-4B base model (e.g., pooled mIoU 81.74→85.78, Acc@0.7 82.51→86.76) and over strong baselines under a unified evaluation protocol. Component ablations attribute the gain to SFT, the box-marked teacher, and IoU-aware weighting.

Significance. If the central claim holds, the paper offers a simple and practically attractive way to exploit ground-truth boxes as privileged visual guidance during training without altering the deployment-time interface. The idea of using a box-overlay plus text hint as a conservative privileged condition, and of weighting distillation by geometrically meaningful token-level factors, is conceptually clear and well motivated. The manuscript is also commendably thorough in documenting the evaluation protocol, the split-level results, and the exact ablation configurations, which makes the reported comparisons reproducible in principle. However, the unique contribution—privileged visual guidance beyond coordinate labels—currently rests on small incremental gains reported as single-run point estimates, with the key temperature hyperparameter selected on the evaluation benchmark. The statistical evidence for the mechanism is therefore not yet at the level needed to support the paper's headline claim.

major comments (3)
  1. [§4.3, Table 7, Appendix E] The central claims about the privileged box teacher and IoU-aware weighting rest on very small single-run differences: B→D (box teacher over original teacher) is +0.65 Acc@0.7, D→E (token weighting) is +0.27, and E→Full is +0.21. For fine-tuning a 4B VLM, these differences are within plausible run-to-run variance. Moreover, the full model's τ_r=0.5 was selected by comparing τ_r=0.5 vs 1.0 on the evaluation benchmark itself (Appendix E, rows F and Full), which introduces optimistic selection bias. The paper needs multiple seeds or a proper significance test, and hyperparameter selection on a validation split (or nested evaluation), before the claim that the box overlay and IoU weighting contribute beyond SFT is supported.
  2. [§3.1, §3.2, Appendix C] The teacher conditions on the ground-truth box overlay and a privileged text hint, while the student does not. The paper acknowledges this conditional mismatch ('The teacher distribution is informative, but it is not identical...') but provides no quantitative check of how large the mismatch is or whether the privileged evidence transfers to the student's inference-time inputs. The argument that the box overlay is a 'conservative privileged hint' is qualitative. Since the entire novel contribution is this privileged visual guidance, the paper should include at least a per-example teacher-student distribution divergence analysis, or an experiment that varies the hint strength / removes the box to show the student behavior is not being pushed toward predictions inconsistent with its actual inputs.
  3. [§4.3, Table 7] The row labeled 'Original teacher' (Variant B) shows Acc@0.5 dropping from 90.62 (SFT-only) to 89.80 while Acc@0.7 rises from 85.32 to 85.63. This non-monotonic behavior is not discussed. It may indicate a shift in the precision/recall tradeoff rather than a uniform improvement, and could affect the interpretation of the subsequent box-teacher comparison. The paper should explain this pattern or at least acknowledge it.
minor comments (5)
  1. [§4.3 / Appendix E] Variant naming is inconsistent between the main text (Table 2 rows 'Original teacher', 'Box teacher', 'Token weighting') and Appendix E/Figure 3 labels ('Orig.', 'Box', 'Weight'). Please harmonize the naming to avoid confusion.
  2. [§3.4] The coordinate error δ_k is defined in normalized coordinate units, but the token-level weights are applied to digit tokens. The text does not specify how δ_k interacts with the digit-position factor; for example, a coordinate error of 40 in the hundreds digit versus the ones digit should presumably have different geometric impact. Clarify the scaling or the implicit normalization.
  3. [Figure 5 / Table 4] The small-object group contains only 375 examples, yet the figure presents it on the same visual scale as the other groups. Please add the sample size to the figure caption or annotate it directly, since the small-object gains are the least statistically stable.
  4. [Abstract / Title] The acronym appears as both 'IoUPD' (abstract) and 'IOU-PD' (elsewhere, including the title). Please use a single consistent spelling throughout.
  5. [Appendix G.1] No code or trained checkpoints are provided, and the artifact section only promises that 'any released code' will follow licenses. Given the small incremental gains, releasing code and checkpoints would substantially strengthen reproducibility and is presumably expected for a training-method paper.

Circularity Check

0 steps flagged

No circular derivation: the privileged teacher is frozen and detached, and the reported gains are empirical results on held-out evaluation splits.

full rationale

The paper's training chain is not circular. The student is trained with L = L_kd + α L_sft, where L_sft is ordinary coordinate-label likelihood and L_kd is a stop-gradient KL against a frozen teacher conditioned on a box-marked image and a privileged hint that are absent at inference. The IoU-aware weights r(u), c_t, d_t, a_t, and e_t are computed from the current student sample's overlap with the ground-truth box and from teacher/student probabilities; they define a training objective, not a substitute for the evaluation metric. The teacher is detached and not updated, so the student is not simply regressing to a fitted version of the test target. Evaluation is performed on pooled held-out splits with no teacher, no box overlay, and no privileged hint, so the claimed improvements are external empirical outcomes rather than results forced by construction. The only methodological concern is that τ_r = 0.5 was selected by comparing ablated variants on the evaluation benchmark; this is a hyperparameter-selection / statistical-risk issue, not a definitional equivalence, and the central box-teacher versus original-teacher comparison does not depend on that choice. No load-bearing self-citation or imported uniqueness theorem is present.

Axiom & Free-Parameter Ledger

6 free parameters · 5 axioms · 2 invented entities

The central claim depends on hand-set weighting hyperparameters and on the transferability of the privileged teacher signal. No new physical entities are introduced; the teacher branch and hint are method components with only in-paper evidence. The most consequential choices are τ_r (selected on the evaluation benchmark) and the multiplicative weight factorization.

free parameters (6)
  • SFT anchor coefficient α = 1.0
    Chosen by hand; balances distillation vs supervised fine-tuning. Variant C uses α=0, showing the anchor matters.
  • Sample-level IoU temperature τ_r = 0.5
    Selected over τ_r=1.0 using the same evaluation benchmark (Figure 3/Table 7); controls exponential weighting by IoU.
  • Coordinate-error softmax temperature τ_c = 1.0
    Chosen by hand; controls softmax weighting over the four coordinate errors.
  • Digit-position factor λ_d = 0.5
    Chosen by hand; scales preference for more significant decimal digits.
  • Teacher-student agreement sigmoid gate β = 3.0
    Chosen by hand; controls how strongly teacher-student agreement modulates token weight.
  • Teacher-confidence entropy temperature τ_e = 1.0
    Chosen by hand; scales the exponential teacher-entropy confidence factor.
axioms (5)
  • domain assumption Privileged box-marked teacher provides a distillation target that improves the student despite teacher and student conditioning on different inputs.
    Stated in §3.1 and defended qualitatively in §C as a 'conservative privileged hint'. No quantitative measure of teacher-student distributional mismatch is provided; if the teacher relies on evidence the student cannot see, the KL term may not transfer.
  • ad hoc to paper The multiplicative weighting w_t = r(u)·c_t·d_t·a_t·e_t is a valid proxy for the contribution of each token to final grounding IoU.
    Introduced in §3.4 with hand-set hyperparameters; no derivation or independent validation that this factorization, rather than uniform weighting, is appropriate for all coordinate formats.
  • domain assumption The 300k grounding examples used for training do not overlap with the five reported evaluation splits.
    §4.1 calls the splits 'held-out' but does not specify exactly how the 300k examples were constructed or whether RefCOCO/RefCOCO+/RefCOCOg training or validation splits were mixed; overlap would inflate all gains.
  • domain assumption A frozen copy of the same base checkpoint is a useful teacher for this distillation scheme.
    Used in §3.2/A.2; supported only by the 'Original teacher' ablation in Table 7, not by comparison to an independently trained or larger teacher.
  • domain assumption The unified parsing/canonicalization protocol evaluates all models fairly despite differing output formats.
    Appendix B specifies a shared parser and fallback rule, but a single fallback for malformed output can favor models that rarely emit parseable strings; no error analysis is provided.
invented entities (2)
  • Green-rectangle box overlay for teacher input no independent evidence
    purpose: Gives the frozen teacher explicit visual localization of the ground-truth region during training only.
    Supported only by the paper's internal ablation (D vs B in Table 7); no external or formal evidence that this specific overlay transfers.
  • Teacher-side text hint 'The answer is located within the green rectangle' no independent evidence
    purpose: Anchors teacher attention to the marked region without revealing coordinates.
    Chosen by design; §C compares it to other hints only qualitatively, not with an independent quantitative test.

pith-pipeline@v1.3.0-alltime-deepseek · 17180 in / 16995 out tokens · 130750 ms · 2026-08-01T22:28:18.523246+00:00 · methodology

0 comments
read the original abstract

Visual grounding with multimodal large language models is commonly formulated as autoregressive coordinate generation, where a model outputs bounding-box coordinates as text given an image and a referring-expression prompt. While this interface is simple and compatible with instruction following, it introduces a mismatch between training and evaluation: training optimizes token-level likelihood over coordinate strings, whereas grounding quality is measured by geometric overlap. We propose IoUPD, an IoU-aware privileged distillation method for coordinate-generating multimodal large language models. IoUPD uses ground-truth boxes not only as coordinate targets, but also as privileged training-time guidance. During training, the student receives the original image and prompt, while a frozen teacher receives a box-marked image and an augmented prompt that indicates the marked region. The student is trained with a supervised fine-tuning anchor and a privileged distillation loss whose token weights reflect both geometric importance and teacher reliability. At inference time, IoUPD requires no box overlay, privileged hint, teacher branch, or additional prediction module. Experiments on standard referring-expression grounding benchmarks show consistent region-level improvements over strong coordinate-generating baselines, demonstrating that ground-truth boxes can provide useful privileged guidance beyond serving as coordinate labels.

Figures

Figures reproduced from arXiv: 2607.15732 by Dongming Zhang, Hao Wu, Jian Xue, Ke Lu, Xiuyuan Zhu, Zijin Du.

Figure 1
Figure 1. Figure 1: Overview of IOU-PD. Ground-truth boxes are used not only as coordinate targets, but also to construct privileged teacher inputs during training. The student receives the original image and original referring-expression prompt, while the teacher receives a box-marked image and an augmented prompt that indicates the marked region. The training objective combines an SFT anchor with IoU-aware privileged distil… view at source ↗
Figure 2
Figure 2. Figure 2: Visualization of IoU-aware token weighting. tokens. [PITH_FULL_IMAGE:figures/full_fig_p005_2.png] view at source ↗
Figure 3
Figure 3. Figure 3: Component ablations of IOU-PD. The upper panel shows the enabled training components, and the lower panel reports Acc@0.5 and Acc@0.7 gains over the base model under the main setting. this setting, the teacher receives the same original image as the student. It improves Acc@.7 from 85.32 to 85.63 over SFT, but the gain is smaller than using a box-marked teacher. Replacing the original teacher with the priv… view at source ↗
Figure 4
Figure 4. Figure 4: Scaling ablations under different data sizes, epoch budgets, and final model settings. The curves show [PITH_FULL_IMAGE:figures/full_fig_p007_4.png] view at source ↗
Figure 5
Figure 5. Figure 5: IOU-PD improves grounding across ob￾ject sizes. Examples are grouped by the ground-truth bounding-box area in the normalized coordinate space: small (< 5%), medium (5%−10%), and large (> 10%). Points report P@0.5 and P@0.7 for the base model and the IOU-PD model, with orange segments and labels indicating absolute gains. only improving easy large-object cases, but also helps the model localize more size-se… view at source ↗
Figure 6
Figure 6. Figure 6: further examines how the improvement changes across IoU thresholds. Compared with the 4B base model, IOU-PD improves P@0.5 by 2.98 points, P@0.7 by 4.25 points, P@0.9 by 7.75 points, and P@0.95 by 12.84 points. The IoU dis￾tribution shows the same trend: predictions be￾low 0.5 IoU decrease from 11.4% to 8.4%, while predictions above 0.95 IoU increase from 35.7% to 48.5%. These results suggest that the priv… view at source ↗
Figure 7
Figure 7. Figure 7: Comparison of different privileged visual [PITH_FULL_IMAGE:figures/full_fig_p012_7.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

136 extracted references · 2 canonical work pages

  1. [1]

    Aho and Jeffrey D

    Alfred V. Aho and Jeffrey D. Ullman , title =. 1972

    1972

  2. [2]

    Publications Manual , year = "1983", publisher =

    1983

  3. [3]

    Chandra and Dexter C

    Ashok K. Chandra and Dexter C. Kozen and Larry J. Stockmeyer , year = "1981", title =. doi:10.1145/322234.322243

    arXiv 1981

  4. [4]

    Scalable training of

    Andrew, Galen and Gao, Jianfeng , booktitle=. Scalable training of

  5. [5]

    Dan Gusfield , title =. 1997

    1997

  6. [6]

    Tetreault , title =

    Mohammad Sadegh Rasooli and Joel R. Tetreault , title =. Computing Research Repository , volume =. 2015 , url =

    2015

  7. [7]

    A Framework for Learning Predictive Structures from Multiple Tasks and Unlabeled Data , Volume =

    Ando, Rie Kubota and Zhang, Tong , Issn =. A Framework for Learning Predictive Structures from Multiple Tasks and Unlabeled Data , Volume =. Journal of Machine Learning Research , Month = dec, Numpages =

  8. [8]

    arXiv preprint arXiv:2306.15195 , year=

    Shikra: Unleashing multimodal llm's referential dialogue magic , author=. arXiv preprint arXiv:2306.15195 , year=

    Pith/arXiv arXiv

  9. [9]

    International Conference on Learning Representations , year=

    Pix2seq: A Language Modeling Framework for Object Detection , author=. International Conference on Learning Representations , year=

  10. [10]

    arXiv preprint arXiv:2306.14824 , year=

    Kosmos-2: Grounding multimodal large language models to the world , author=. arXiv preprint arXiv:2306.14824 , year=

    Pith/arXiv arXiv

  11. [11]

    Advances in Neural Information Processing Systems , volume=

    Cogvlm: Visual expert for pretrained language models , author=. Advances in Neural Information Processing Systems , volume=

  12. [12]

    Ferret: Refer and Ground Anything Anywhere at Any Granularity , author=

  13. [13]

    European Conference on Computer Vision , pages=

    Griffon: Spelling out all object locations at any granularity with large language models , author=. European Conference on Computer Vision , pages=. 2024 , organization=

    2024

  14. [14]

    3D Gaussian Splatting for Real-Time Radiance Field Rendering , journal =

    Kerbl, Bernhard and Kopanas, Georgios and Leimk. 3D Gaussian Splatting for Real-Time Radiance Field Rendering , journal =

  15. [15]

    Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition (CVPR) , month =

    Wu, Guanjun and Yi, Taoran and Fang, Jiemin and Xie, Lingxi and Zhang, Xiaopeng and Wei, Wei and Liu, Wenyu and Tian, Qi and Wang, Xinggang , title =. Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition (CVPR) , month =. 2024 , pages =

    2024

  16. [16]

    European Conference on Computer Vision , pages=

    Citygaussian: Real-time high-quality large-scale scene rendering with gaussians , author=. European Conference on Computer Vision , pages=. 2025 , organization=

    2025

  17. [17]

    arXiv preprint arXiv:2404.14249 , year=

    CLIP-GS: CLIP-Informed Gaussian Splatting for Real-time and View-consistent 3D Semantic Understanding , author=. arXiv preprint arXiv:2404.14249 , year=

    Pith/arXiv arXiv

  18. [18]

    ECCV , year=

    CompGS: Smaller and Faster Gaussian Splatting with Vector Quantization , author=. ECCV , year=

  19. [19]

    Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition , pages=

    Feature 3dgs: Supercharging 3d gaussian splatting to enable distilled feature fields , author=. Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition , pages=

  20. [20]

    European Conference on Computer Vision (ECCV) , year=

    Language-Driven Physics-Based Scene Synthesis and Editing via Feature Splatting , author=. European Conference on Computer Vision (ECCV) , year=

  21. [21]

    ECCV , year=

    Gaussian Grouping: Segment and Edit Anything in 3D Scenes , author=. ECCV , year=

  22. [22]

    2024 , booktitle =

    Fang, Jiemin and Wang, Junjie and Zhang, Xiaopeng and Xie, Lingxi and Tian, Qi , title =. 2024 , booktitle =

    2024

  23. [23]

    arXiv:2304.02643 , year=

    Segment Anything , author=. arXiv:2304.02643 , year=

    Pith/arXiv arXiv

  24. [24]

    2024 , eprint=

    SAM 2: Segment Anything in Images and Videos , author=. 2024 , eprint=

    2024

  25. [25]

    arXiv preprint arXiv:2303.05499 , year=

    Grounding dino: Marrying dino with grounded pre-training for open-set object detection , author=. arXiv preprint arXiv:2303.05499 , year=

    Pith/arXiv arXiv

  26. [26]

    2024 , eprint=

    Grounded SAM: Assembling Open-World Models for Diverse Visual Tasks , author=. 2024 , eprint=

    2024

  27. [27]

    Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition (CVPR) , month =

    Zhou, Hongyu and Shao, Jiahao and Xu, Lu and Bai, Dongfeng and Qiu, Weichao and Liu, Bingbing and Wang, Yue and Geiger, Andreas and Liao, Yiyi , title =. Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition (CVPR) , month =. 2024 , pages =

    2024

  28. [28]

    Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition (CVPR) , month =

    Qin, Minghan and Li, Wanhua and Zhou, Jiawei and Wang, Haoqian and Pfister, Hanspeter , title =. Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition (CVPR) , month =. 2024 , pages =

    2024

  29. [29]

    Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition (CVPR) , month =

    Shi, Jin-Chuan and Wang, Miao and Duan, Hao-Bin and Guan, Shao-Hua , title =. Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition (CVPR) , month =. 2024 , pages =

    2024

  30. [30]

    International Conference on Computer Vision (ICCV) , year =

    Kerr, Justin* and Kim, Chung Min* and Goldberg, Ken and Kanazawa, Angjoo and Tancik, Matthew , title =. International Conference on Computer Vision (ICCV) , year =

  31. [31]

    Proceedings of the Winter Conference on Applications of Computer Vision (WACV) , month =

    Chacko, Rohan and Haeni, Nicolai and Khaliullin, Eldar and Sun, Lin and Lee, Douglas , title =. Proceedings of the Winter Conference on Applications of Computer Vision (WACV) , month =. 2025 , pages =

    2025

  32. [32]

    The Thirty-eighth Annual Conference on Neural Information Processing Systems , year=

    OpenGaussian: Towards Point-Level 3D Gaussian-based Open Vocabulary Understanding , author=. The Thirty-eighth Annual Conference on Neural Information Processing Systems , year=

  33. [33]

    ECCV , year =

    Pixel-GS: Density Control with Pixel-aware Gradient for 3D Gaussian Splatting , author =. ECCV , year =

  34. [34]

    arXiv preprint arXiv:2411.19290 , year=

    SADG: Segment Any Dynamic Gaussian Without Object Trackers , author=. arXiv preprint arXiv:2411.19290 , year=

    arXiv

  35. [35]

    2023 , journal=

    Segment Any 3D Gaussians , author=. 2023 , journal=

    2023

  36. [36]

    2024 , eprint=

    Semantic Gaussians: Open-Vocabulary Scene Understanding with 3D Gaussian Splatting , author=. 2024 , eprint=

    2024

  37. [37]

    European Conference on Computer Vision , year=

    Mini-Splatting: Representing Scenes with a Constrained Number of Gaussians , author=. European Conference on Computer Vision , year=

  38. [38]

    2024 , journal =

    TRIPS: Trilinear Point Splatting for Real-Time Radiance Field Rendering , author=. 2024 , journal =

    2024

  39. [39]

    arXiv preprint arXiv:2403.19615 , year=

    SA-GS: Scale-Adaptive Gaussian Splatting for Training-Free Anti-Aliasing , author=. arXiv preprint arXiv:2403.19615 , year=

    Pith/arXiv arXiv

  40. [40]

    Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition , pages=

    Multi-scale 3d gaussian splatting for anti-aliased rendering , author=. Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition , pages=

  41. [41]

    2025 , volume =

    Qiu, Shi and Xie, Binzhu and Liu, Qixuan and Heng, Pheng-Ann , booktitle =. 2025 , volume =. doi:10.1109/AIxVR63409.2025.00039 , publisher =

    arXiv 2025

  42. [42]

    Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition , pages=

    Drivinggaussian: Composite gaussian splatting for surrounding dynamic autonomous driving scenes , author=. Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition , pages=

  43. [43]

    The Thirteenth International Conference on Learning Representations , year=

    OmniRe: Omni Urban Scene Reconstruction , author=. The Thirteenth International Conference on Learning Representations , year=

  44. [44]

    Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition (CVPR) , month =

    Yang, Ziyi and Gao, Xinyu and Zhou, Wen and Jiao, Shaohui and Zhang, Yuqing and Jin, Xiaogang , title =. Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition (CVPR) , month =. 2024 , pages =

    2024

  45. [45]

    Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition (CVPR) , month =

    Huang, Yi-Hua and Sun, Yang-Tian and Yang, Ziyi and Lyu, Xiaoyang and Cao, Yan-Pei and Qi, Xiaojuan , title =. Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition (CVPR) , month =. 2024 , pages =

    2024

  46. [46]

    Proceedings of the IEEE Conference on Computer Vision and Pattern Recognition (CVPR) , year=

    PointGroup: Dual-Set Point Grouping for 3D Instance Segmentation , author=. Proceedings of the IEEE Conference on Computer Vision and Pattern Recognition (CVPR) , year=

  47. [47]

    and Millis, Bryan A

    Zhao, Mengyang and Liu, Quan and Jha, Aadarsh and Deng, Ruining and Yao, Tianyuan and Mahadevan-Jansen, Anita and Tyska, Matthew J. and Millis, Bryan A. and Huo, Yuankai , title =. 2021 , isbn =. doi:10.1007/978-3-030-87589-3_45 , booktitle =

    work page doi:10.1007/978-3-030-87589-3_45 2021

  48. [48]

    Proceedings of the IEEE/CVF International Conference on Computer Vision (ICCV) , month =

    Kerr, Justin and Kim, Chung Min and Goldberg, Ken and Kanazawa, Angjoo and Tancik, Matthew , title =. Proceedings of the IEEE/CVF International Conference on Computer Vision (ICCV) , month =. 2023 , pages =

    2023

  49. [49]

    European Conference on Computer Vision (ECCV) , year =

    Anpei Chen and Zexiang Xu and Andreas Geiger and Jingyi Yu and Hao Su , title =. European Conference on Computer Vision (ECCV) , year =

  50. [50]

    Thirty-seventh Conference on Neural Information Processing Systems , year=

    Weakly Supervised 3D Open-vocabulary Segmentation , author=. Thirty-seventh Conference on Neural Information Processing Systems , year=

  51. [51]

    Proceedings of the 38th International Conference on Machine Learning , pages =

    Learning Transferable Visual Models From Natural Language Supervision , author =. Proceedings of the 38th International Conference on Machine Learning , pages =. 2021 , editor =

    2021

  52. [52]

    and Tancik, Matthew and Barron, Jonathan T

    Mildenhall, Ben and Srinivasan, Pratul P. and Tancik, Matthew and Barron, Jonathan T. and Ramamoorthi, Ravi and Ng, Ren , title =. 2021 , issue_date =. doi:10.1145/3503250 , journal =

    doi:10.1145/3503250 2021

  53. [53]

    ACM Trans

    Thomas M\""uller and Alex Evans and Christoph Schied and Alexander Keller , title =. ACM Trans. Graph. , issue_date =. 2022 , pages =. doi:10.1145/3528223.3530127 , publisher =

    arXiv 2022

  54. [54]

    and Kowalski, Marek and Johnson, Matthew and Shotton, Jamie and Valentin, Julien , title =

    Garbin, Stephan J. and Kowalski, Marek and Johnson, Matthew and Shotton, Jamie and Valentin, Julien , title =. Proceedings of the IEEE/CVF International Conference on Computer Vision (ICCV) , month =. 2021 , pages =

    2021

  55. [55]

    2022 , booktitle=

    Plenoxels: Radiance Fields without Neural Networks , author=. 2022 , booktitle=

    2022

  56. [56]

    2025 , issn =

    DSEM-NeRF: Multimodal feature fusion and global–local attention for enhanced 3D scene reconstruction , journal =. 2025 , issn =. doi:https://doi.org/10.1016/j.inffus.2024.102752 , author =

    arXiv 2025

  57. [57]

    Pumarola, Albert and Corona, Enric and Pons-Moll, Gerard and Moreno-Noguer, Francesc , booktitle=

  58. [58]

    2024 , isbn =

    Lin, Youtian , title =. 2024 , isbn =. doi:10.1609/aaai.v38i4.28138 , booktitle =

    work page doi:10.1609/aaai.v38i4.28138 2024

  59. [59]

    ICCV , year=

    Mip-NeRF: A Multiscale Representation for Anti-Aliasing Neural Radiance Fields , author=. ICCV , year=

  60. [60]

    Conference on Computer Vision and Pattern Recognition (CVPR) , year =

    Yu, Zehao and Chen, Anpei and Huang, Binbin and Sattler, Torsten and Geiger, Andreas , title =. Conference on Computer Vision and Pattern Recognition (CVPR) , year =

  61. [61]

    CVPR , year=

    SuGaR: Surface-Aligned Gaussian Splatting for Efficient 3D Mesh Reconstruction and High-Quality Mesh Rendering , author=. CVPR , year=

  62. [62]

    arXiv preprint arXiv:2403.06908 , year=

    FreGS: 3D Gaussian Splatting with Progressive Frequency Regularization , author=. arXiv preprint arXiv:2403.06908 , year=

    Pith/arXiv arXiv

  63. [63]

    ECCV , year=

    Street Gaussians for Modeling Dynamic Urban Scenes , author=. ECCV , year=

  64. [64]

    CVPR , year =

    VastGaussian: Vast 3D Gaussians for Large Scene Reconstruction , author =. CVPR , year =

  65. [65]

    2025 , eprint=

    A Hierarchical Compression Technique for 3D Gaussian Splatting Compression , author=. 2025 , eprint=

    2025

  66. [66]

    arXiv preprint arXiv:2403.17898 , year=

    Octree-gs: Towards consistent real-time rendering with lod-structured 3d gaussians , author=. arXiv preprint arXiv:2403.17898 , year=

    Pith/arXiv arXiv

  67. [67]

    Proceedings of the IEEE/CVF International Conference on Computer Vision , pages=

    Matrixcity: A large-scale city dataset for city-scale neural rendering and beyond , author=. Proceedings of the IEEE/CVF International Conference on Computer Vision , pages=

  68. [68]

    Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition (CVPR) , month =

    Turki, Haithem and Ramanan, Deva and Satyanarayanan, Mahadev , title =. Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition (CVPR) , month =. 2022 , pages =

    2022

  69. [69]

    A density-based algorithm for discovering clusters in large spatial databases with noise , year =

    Ester, Martin and Kriegel, Hans-Peter and Sander, J\"". A density-based algorithm for discovering clusters in large spatial databases with noise , year =. Proceedings of the Second International Conference on Knowledge Discovery and Data Mining , pages =

  70. [70]

    doi:10.21105/joss.00205 , year =

    Leland McInnes and John Healy and Steve Astels , title =. doi:10.21105/joss.00205 , year =

    doi:10.21105/joss.00205

  71. [71]

    Xu and Jun-Mei Song and Mingchuan Zhang and Y

    Zhihong Shao and Peiyi Wang and Qihao Zhu and R. Xu and Jun-Mei Song and Mingchuan Zhang and Y. K. Li and Yu Wu and Daya Guo , booktitle =. ArXiv , title =

  72. [72]

    Nature , volume=

    Deepseek-r1 incentivizes reasoning in llms through reinforcement learning , author=. Nature , volume=. 2025 , publisher=

    2025

  73. [73]

    Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition (CVPR) , month =

    Alvar, Saeed Ranjbar and Singh, Gursimran and Akbari, Mohammad and Zhang, Yong , title =. Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition (CVPR) , month =. 2025 , pages =

    2025

  74. [74]

    European conference on computer vision , pages=

    Mmbench: Is your multi-modal model an all-around player? , author=. European conference on computer vision , pages=. 2024 , organization=

    2024

  75. [75]

    2024 , eprint=

    MME: A Comprehensive Evaluation Benchmark for Multimodal Large Language Models , author=. 2024 , eprint=

    2024

  76. [76]

    arXiv preprint arXiv:2506.10967 , year=

    Beyond Attention or Similarity: Maximizing Conditional Diversity for Token Pruning in MLLMs , author=. arXiv preprint arXiv:2506.10967 , year=

    Pith/arXiv arXiv

  77. [77]

    Proceedings of the IEEE/CVF Winter Conference on Applications of Computer Vision , year=

    FairFace: Face Attribute Dataset for Balanced Race, Gender, and Age for Bias Measurement and Mitigation , author=. Proceedings of the IEEE/CVF Winter Conference on Applications of Computer Vision , year=

  78. [78]

    arXiv preprint arXiv:2411.19930 , year=

    On Domain-Adaptive Post-Training for Multimodal Large Language Models , author=. arXiv preprint arXiv:2411.19930 , year=

    Pith/arXiv arXiv

  79. [79]

    arXiv preprint arXiv:2412.05271 , year=

    Expanding Performance Boundaries of Open-Source Multimodal Models with Model, Data, and Test-Time Scaling , author=. arXiv preprint arXiv:2412.05271 , year=

    Pith/arXiv arXiv

  80. [80]

    arXiv preprint arXiv:2411.10442 , year=

    Enhancing the Reasoning Ability of Multimodal Large Language Models via Mixed Preference Optimization , author=. arXiv preprint arXiv:2411.10442 , year=

    Pith/arXiv arXiv

Showing first 80 references.