Pith. sign in

REVIEW 4 major objections 6 minor 60 references

A feedforward network trained purely on 2D supervision can match or beat task-specific 3D models on segmentation and correspondence, and stays accurate under arbitrary rotations.

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 05:02 UTC pith:775ZERCP

load-bearing objection Solid distillation pipeline, but the 'all you need' claim needs a disjointness statement and more honest comparisons before it earns the title. the 4 major comments →

arxiv 2607.27592 v1 pith:775ZERCP submitted 2026-07-30 cs.CV cs.GR

MeshFM: 2D Features Are All You Need for 3D Shape Understanding

Jinfan Zhou , Richard Liu , Itai Lang , Rana Hanocka This is my paper
classification cs.CV cs.GR
keywords MeshFM2D feature distillationfeature field3D shape understandingpart segmentationdense correspondencerotation robustnesszero-shot transfer
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 sets out to show that task-specific 3D networks are not necessary for 3D shape understanding: the apparent need for dedicated heads is, in the authors' view, an artifact of imperfect feature distillation. It introduces MeshFM, a feedforward network that predicts a general-purpose, rotation-robust 3D feature field for a mesh in one forward pass, trained only with 2D supervision. The enabling steps are a two-stage distillation—first fitting a clean per-shape teacher field from multi-view 2D features, with a segmentation-based correction that removes patch-induced 'feature bleeding' across part boundaries, then regressing that field with a point-cloud network under full rotation augmentation. If correct, the claim implies that a single set of 2D-derived features can replace specialized heads for segmentation, correspondence, classification, and deformation, and that 3D annotation is not a bottleneck for these tasks.

Core claim

The paper's central discovery is that distilled 2D foundation-model features, once corrected for aliasing and passed through a two-stage training procedure, are sufficient to match or beat task-specific 3D-supervised methods on part segmentation, dense correspondence, classification, and mesh deformation. The enabling correction is a segmentation-based feature-aliasing step: within each segment of an image-based segmentation model, pixels whose features deviate too far from the segment's median feature are replaced by that median, suppressing patch-level 'feature bleeding' across semantic boundaries. After the corrected teacher field is optimized, a feedforward network can regress it directl

What carries the argument

The load-bearing object is the corrected teacher feature field: a per-shape neural field fitted by back-projecting multi-view 2D foundation-model features onto the mesh, then denoised by replacing outlier pixels within each segmentation mask with that mask's median feature. This clean field supervises a feedforward predictor (point-voxel encoder with triplane representation and transformer) that produces rotation-robust features in a single pass.

Load-bearing premise

The correction assumes that pixels inside a single segmentation mask should all carry nearly the same feature and that large deviations are patch-bleeding noise; if the mask is imprecise or the bleeding contaminates the segment median itself, the corrected teacher field injects wrong features into every downstream prediction.

What would settle it

Run the pipeline with the segmentation-based correction replaced by a comparable denoiser that does not depend on mask boundaries (for example, median filtering in feature space over local neighborhoods), or perturb the segmentation masks slightly on shapes with thin parts. If downstream segmentation and correspondence do not degrade when the masks are inaccurate, the specific mask-median mechanism is not the decisive factor; if the method collapses under corrupted masks, the assumption is confirmed as load-bearing.

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

If this is right

  • On class-agnostic part segmentation, the general-purpose features reach or slightly exceed a leading segmentation-specialized method on the original test sets, and clearly outperform it when shapes are randomly rotated.
  • On dense correspondence, the method matches task-specific correspondence methods on canonical poses and remains accurate under SO(3) rotation, where the baselines degrade sharply.
  • The same features serve segmentation, correspondence, classification, and deformation zero-shot, whereas the task-specific baselines do not transfer across tasks.
  • The segmentation-based aliasing correction improves teacher-field quality consistently, raising instance-segmentation mean IoU by about three points on average and improving correspondence accuracy.
  • The two-stage design beats directly training the feedforward network on back-projected 2D features, indicating that teacher-field optimization filters multi-view inconsistency.

Where Pith is reading between the lines

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

  • If this claim holds, progress in 3D understanding could piggyback on progress in 2D foundation models: each new 2D model can be plugged into the same two-stage pipeline, a possibility the paper's foundation-model ablation already hints at.
  • The acknowledged left-right confusion in correspondence suggests a concrete extension: inject a weak geometric or positional signal into the corrected teacher field, which could remove the main failure mode while keeping the features annotation-free.
  • The fact that rotation robustness comes purely from augmentation suggests task-specific 3D models trained only on canonical poses could adopt the same training recipe, though their dependence on 3D-supervised or mask-based signals may make it harder.
  • A natural stress test of the thesis is scene-level understanding, where occlusion and multi-object fields introduce ambiguities that single-object distillation does not; the paper's per-object results leave this open.

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. The paper proposes MeshFM, a two-stage framework for learning general-purpose 3D shape features from 2D foundation models. In Stage 1, per-shape neural feature fields are optimized by distilling DINOv2 (or other) features from multi-view renderings, with a SAM-based aliasing correction that replaces outlier pixels in each segment by the segment median. In Stage 2, a feedforward PVCNN+triplane-transformer network is trained to regress these teacher fields, with SO(3) rotation augmentation. The resulting features are then applied zero-shot to part segmentation, dense correspondence, classification, and deformation, and compared against PartField, Diff3F, and DenseMatcher. The central claim is that properly distilled 2D features are sufficient for near-SOTA 3D understanding without task-specific heads or 3D supervision.

Significance. If validated, the paper would support a strong and useful thesis: that 2D foundation features, when distilled with boundary-aware correction, can serve as a general-purpose representation for diverse 3D tasks, eliminating the need for task-specific 3D supervision. The paper is well positioned in the literature and includes extensive experiments, a fair common postprocessing protocol, ablations (SAM correction, foundation-model choice, two-stage design), inference-time analysis, and cost breakdowns. These are genuine strengths. However, the current evaluation has critical gaps: no demonstration of train/test disjointness for Objaverse-derived benchmarks, no error bars or multiple seeds, and a clear case where the paper's reported numbers contradict the 'on par' wording (TOSCA). These issues must be resolved before the central claim can be accepted.

major comments (4)
  1. [S2.1, §4] The paper never states that the 130k-shape Objaverse training subset (S2.1) is disjoint from the evaluation benchmarks. PartObjaverse-Tiny (Table 1, S1) and DenseCorr3D (Table 3) are derived from Objaverse/Objaverse-XL, so train/test overlap is plausible. Because Stage 2 regresses the exact teacher fields produced in Stage 1, any test shape appearing in training could be memorized rather than generalized to. A clear statement of disjointness, or a retrained evaluation on a held-out split, is a necessary condition for the paper's headline claim of zero-shot transfer to unseen 3D shapes.
  2. [Tables 1, 2, S1, S2] No error bars, confidence intervals, or multiple seeds are reported. The reported gains over PartField are very small (e.g., Table 1: 0.549 vs 0.542; Table 2: 0.520 vs 0.517), and in Table S1 MeshFM is actually below PartField on original PartObjaverse-Tiny instance segmentation (0.598 vs 0.652) despite the text saying 'slightly underperforms'. Without variance estimates, the claims of statistical superiority or parity are not supported. The authors should report mean and standard deviation over at least 3 runs, or otherwise justify that the differences are not noise.
  3. [Table 4, §4.2] On the original (non-rotated) TOSCA benchmark, MeshFM reports AUC 0.34 and Err 18.8, while Diff3F reports 0.48/17.1 and DenseMatcher 0.50/10.2. The text states 'Our method is on par with state-of-the-art baselines,' but these numbers show a substantial underperformance, especially in AUC. This directly weakens the generality claim in the abstract and conclusion. The authors should either report additional metrics or revise the claim; if the underperformance is due to the functional-map postprocessing, that should be analyzed and disclosed.
  4. [Eq. (1)-(2), §3.1] The SAM aliasing correction assumes that each SAM segment has a unimodal feature distribution and that the median is a reliable representative. If a SAM mask is imprecise, or if patch bleeding corrupts the segment median itself, replacing outliers by the median can inject incorrect features into the teacher field, and every downstream result inherits this error. The ablation (Table 6, S4) shows only modest improvements (e.g., +0.031 mIoU average on PartObjaverse-Tiny; +0.01 AUC on DenseCorr3D), so it is not clear that this mechanism is the 'key enabler' described in Section 1. The authors should test robustness to SAM mask errors, e.g., by perturbing masks or by ablating with an alternative boundary correction.
minor comments (6)
  1. [§1] Typo: 'Championingthishypothesis' should be 'Championing this hypothesis'.
  2. [Figure 3] The labels 'Original Distillation' and 'Corrected Distillation' are small; consider enlarging or adding arrows to clarify the comparison.
  3. [§3.1] The threshold tau=1 is stated without sensitivity analysis. Since this is a hyperparameter of the proposed correction, a brief study across tau values would strengthen the paper.
  4. [S2.3] The sentence 'This preprocessing step is necessary regardless whether Stage 1 is incorporated' is unclear: for the Stage-2-only ablation, is the preprocessing identical? Clarify the comparison.
  5. [§4.3] Manifold40 classification: the paper reports accuracy but does not describe the number of training samples per class for the MLP, or the data split. Please provide details for reproducibility.
  6. [Abstract] The abstract states 'perform on par with methods trained explicitly with 3D supervision' without qualification; Table 4 shows a counterexample. Recommend adding a qualifying phrase such as 'on the evaluated benchmarks except TOSCA' or reworking the claim.

Circularity Check

0 steps flagged

No derivation reduces to its inputs; the two same-group citations are method adoption and related work, not load-bearing.

full rationale

MeshFM's derivation is a two-stage empirical pipeline: Stage 1 fits per-shape 3D teacher fields from multi-view 2D foundation-model features (with SAM-based median filtering, Eq. 1-2), and Stage 2 regresses those frozen fields with an L2 loss (Eq. 3). No equation defines the predicted quantity in terms of the evaluation target. Downstream evaluations (part segmentation mIoU, correspondence AUC/geodesic error, classification accuracy) are measured against external human annotations or held-out benchmark labels, not against the teacher fields used for training, so the fit-then-predict loop is broken by independent ground truth. The two same-group citations (DFD [24] for barycentric distillation/deformation and iSeg [19] for interactive attention) are explicit methodological borrowings and related-work context; they are not invoked to justify the central claim that distilled 2D features generalize. A legitimate experimental concern is that the 130k-shape Objaverse training set (S2.1) is not stated to be disjoint from Objaverse-derived test sets such as PartObjaverse-Tiny, DenseCorr3D, and Manifold40; if overlap exists, generalization numbers could be inflated. That is a benchmark-hygiene/correctness issue rather than circularity, because the paper does not define its predictions in terms of the test labels. Overall, no load-bearing self-referential reduction was found.

Axiom & Free-Parameter Ledger

4 free parameters · 5 axioms · 0 invented entities

The central pipeline relies on empirical assumptions about the 2D foundation models (DINOv2/RADIO/SAM) and about the 3D supervision chain from multi-view backprojection through teacher fields to feedforward regression. There are no invented physical or latent entities; the feature field and triplane representation are standard. The main hand-set knob is the outlier threshold tau=1, plus several omitted hyperparameters such as lambda_reg and the rendering setup.

free parameters (4)
  • tau (SAM outlier threshold) = 1
    Eq. (2): pixels with ||f_u - mu_i||_2 > 1 are replaced by the segment median. The authors state tau=1 produces reasonable results across all experiments, so it is a hand-set global threshold that directly controls teacher-field quality.
  • cluster count k (segmentation) = number of ground-truth labels per shape
    Unsupervised segmentation evaluation sets k to the number of unique GT labels for all methods. This is privileged oracle information at inference and can overstate true zero-shot capability.
  • lambda_reg (functional map regularizer) = not reported
    Eq. (S5): weight balancing feature preservation vs Laplacian commutativity in the correspondence postprocessing. It is a tuned hyperparameter but its value is omitted from the paper.
  • SO(3) augmentation schedule = 365k extra iterations; theta_max ramps to 2pi in first 125k
    Training scheme chosen by hand to induce rotation robustness. No ablation on the schedule is provided, so the contribution of this specific schedule to the rotation-robustness claim is not isolated.
axioms (5)
  • domain assumption DINOv2/RADIO features are semantically rich and spatially localized enough to transfer to 3D tasks.
    Sections 2.1 and 5: the entire teacher signal comes from these 2D features; the paper inherits their limitations, e.g., left/right confusion.
  • domain assumption SAM segments correspond to semantically consistent regions with similar deep features.
    Section 3.1: outlier correction assumes within-mask features are similar and deviations are patch-bleeding artifacts.
  • domain assumption Barycentric backprojection of 2D features yields a valid, view-consistent continuous 3D supervision signal.
    Section 3.1: adopts DFD's barycentric distillation; no error analysis for occlusions, render coverage, or view inconsistency is provided.
  • domain assumption A PVCNN+triplane feedforward network can regress the teacher field on new shapes from point clouds.
    Section 3.2: generalization from sparse point cloud input to a continuous 3D field is assumed without proof or capacity analysis.
  • domain assumption Multi-view renderings cover the mesh and the 2D features agree across views.
    Stage 1 optimization aggregates renderings into one field; inconsistent views are treated as noise, but this consistency is not demonstrated.

pith-pipeline@v1.3.0-daily-deepseek · 16136 in / 14246 out tokens · 144070 ms · 2026-08-01T05:02:29.296768+00:00 · methodology

0 comments
read the original abstract

We present MeshFM, an efficient feedforward framework for extracting rich features from 3D inputs. Our method distills 2D features from visual foundation models into 3D. We train a feedforward network to directly predict 3D features without requiring optimization during inference. The approach utilizes a two-stage training strategy. First, we optimize a feature field in 3D using only 2D feature supervision. Second, we train a network to regress this feature field. The entire procedure requires no 3D annotation, instead relying on the powerful information in 2D foundation models. We demonstrate that our learned features can be immediately applied to downstream tasks, including part segmentation, dense correspondence, and mesh deformation. Extensive experiments show that MeshFM, trained solely with 2D supervision, performs on par with methods trained explicitly with 3D supervision, even without task-specific fine-tuning. Moreover, our model is trained to be robust to extreme rotations of the input objects. Project page: https://threedle.github.io/MeshFM/

Figures

Figures reproduced from arXiv: 2607.27592 by Itai Lang, Jinfan Zhou, Rana Hanocka, Richard Liu.

Figure 1
Figure 1. Figure 1: MeshFM is a feedforward model that computes general-purpose features for 3D shapes. The learned features are discriminative and capture meaningful properties across diverse shapes and complex geometries. Importantly, the features are multi￾purpose, and can be applied in a zero-shot manner for a variety of downstream tasks, such as segmentation, correspondence, and deformation. Abstract. We present MeshFM, … view at source ↗
Figure 2
Figure 2. Figure 2: Training pipeline overview. The training phase of MeshFM includes two stages. In the first stage, we distill the information from a 2D foundation model into 3D feature fields. 2D feature maps are extracted from multi-view renderings of the mesh and projected to the mesh surface, and a neural model learns to map 3D coordinates on the mesh to the reference features. In the second stage, we use the teacher fe… view at source ↗
Figure 3
Figure 3. Figure 3: Feature aliasing correction. We correct the feature aliasing created by patch-based image encoder architectures through fine-grained image segmentation. The segmentation groups define pixel clusters that are assumed to have similar deep fea￾tures. Pixels with features substantially different than the cluster median are replaced by the median. The resulting feature map has much less blurring at semantic bou… view at source ↗
Figure 4
Figure 4. Figure 4: Part segmentation comparison. Our method can segment the mesh without training on any 3D annotations. We can even perform on par with PartField, which is directly trained with 3D segmentation annotation. In cases where our segmentation differs from the ground-truth, they are still semantically meaningful (e.g. decorative ribbons on the dress). aggregated from all the views. Then, a feature field Φ is optim… view at source ↗
Figure 5
Figure 5. Figure 5: Rotation robustness. We visualize our segmentation results for different rotations of the input shape. MeshFM is highly robust to rotations, where the segmen￾tation remains consistent under large rotations in the full SO(3) group. For every segment Mi ∈ Sk , we first compute the representative segment feature µi as the channel-wise median of all feature vectors within the mask: \mathbf {\mu }_i = \operator… view at source ↗
Figure 6
Figure 6. Figure 6: Hierarchical part segmentation. Our method can segment the mesh at an increasing granularity by simply increasing the number of feature clusters (from left to right). Notably, we support a wide range of part counts, distinguish part instances (e.g., the character’s boots), and capture fine-grained shape details (e.g., the band on the rifle stock). 2. Triplane Transformer: The initial triplanes are processe… view at source ↗
Figure 7
Figure 7. Figure 7: Dense correspondence comparison. While other methods fail on these examples, our mesh features are consistent across shapes and enable dense matching between shapes even when shapes are not in a canonical pose and are not aligned. The correspondence is visualized by color transfer over each pair. Notably, the correspon￾dence applies to various domains and different geometries, including household objects a… view at source ↗
Figure 8
Figure 8. Figure 8: Handle-Based Deformation. We show multi-handle deformations using our features and the Deep Feature Deformation framework [24] on shapes from the Part￾Objaverse dataset. Initial control handles and affine transformations are shown in red. Final handle destinations are shown in green. PartObjaverse-Tiny in [PITH_FULL_IMAGE:figures/full_fig_p014_8.png] view at source ↗
Figure 9
Figure 9. Figure 9: Left-right mismatch limitation. Our mesh features sometimes do not dis￾tinguish left and right symmetric shape parts, such as chair legs and animal limbs, leading to correspondence errors. used the popular DINOv2 model [26], which is known to struggle distinguishing left from right [38], as it mainly reasons about semantic meaning and less about relative position. This limitation is evident in the correspo… 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

60 extracted references · 1 canonical work pages

  1. [1]

    In: Proceedings of the IEEE/CVF conference on computer vision and pattern recognition

    Deitke, M., Schwenk, D., Salvador, J., Weihs, L., Michel, O., VanderBilt, E., Schmidt, L., Ehsani, K., Kembhavi, A., Farhadi, A.: Objaverse: A universe of annotated 3d objects. In: Proceedings of the IEEE/CVF conference on computer vision and pattern recognition. pp. 13142--13153 (2023)

    2023

  2. [2]

    In: Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition

    Dutt, N.S., Muralikrishnan, S., Mitra, N.J.: Diffusion 3d features (diff3f): Decorating untextured shapes with distilled semantic features. In: Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition. pp. 4494--4504 (2024)

    2024

  3. [3]

    In: Proceedings of the International Conference on 3D Vision (3DV)

    Lang, I., Ginzburg, D., Avidan, S., Raviv, D.: DPC: Unsupervised Deep Point Correspondence via Cross and Self Construction. In: Proceedings of the International Conference on 3D Vision (3DV). pp. 1442--1451 (2021)

    2021

  4. [4]

    In: Proceedings of the IEEE/CVF International Conference on Computer Vision

    Lin, C., Liu, H., Lin, Q., Bright, Z., Tang, S., He, Y., Liu, M., Zhu, L., Le, C.: Objaverse++: Curated 3d object dataset with quality annotations. In: Proceedings of the IEEE/CVF International Conference on Computer Vision. pp. 6813--6822 (2025)

    2025

  5. [5]

    In: Proceedings of the IEEE/CVF International Conference on Computer Vision

    Liu, M., Uy, M.A., Xiang, D., Su, H., Fidler, S., Sharp, N., Gao, J.: Partfield: Learning 3d feature fields for part segmentation and beyond. In: Proceedings of the IEEE/CVF International Conference on Computer Vision. pp. 9704--9715 (2025)

    2025

  6. [6]

    In: Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition

    Liu, M., Zhu, Y., Cai, H., Han, S., Ling, Z., Porikli, F., Su, H.: Partslip: Low-shot part segmentation for 3d point clouds via pretrained image-language models. In: Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition. pp. 21736--21746 (2023)

    2023

  7. [9]

    arXiv preprint arXiv:2304.07193 , year=

    Dinov2: Learning robust visual features without supervision , author=. arXiv preprint arXiv:2304.07193 , year=

    Pith/arXiv arXiv

  8. [10]

    International conference on machine learning , pages=

    Learning transferable visual models from natural language supervision , author=. International conference on machine learning , pages=. 2021 , organization=

    2021

  9. [11]

    Proceedings of the IEEE/CVF international conference on computer vision , pages=

    Segment anything , author=. Proceedings of the IEEE/CVF international conference on computer vision , pages=

  10. [12]

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

    Diffusion 3d features (diff3f): Decorating untextured shapes with distilled semantic features , author=. Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition , pages=

  11. [13]

    International conference on machine learning , pages=

    Blip: Bootstrapping language-image pre-training for unified vision-language understanding and generation , author=. International conference on machine learning , pages=. 2022 , organization=

    2022

  12. [14]

    International conference on machine learning , pages=

    Blip-2: Bootstrapping language-image pre-training with frozen image encoders and large language models , author=. International conference on machine learning , pages=. 2023 , organization=

    2023

  13. [15]

    Proceedings of the IEEE/CVF international conference on computer vision , pages=

    Sigmoid loss for language image pre-training , author=. Proceedings of the IEEE/CVF international conference on computer vision , pages=

  14. [16]

    arXiv preprint arXiv:2502.14786 , year=

    Siglip 2: Multilingual vision-language encoders with improved semantic understanding, localization, and dense features , author=. arXiv preprint arXiv:2502.14786 , year=

    Pith/arXiv arXiv

  15. [17]

    Proceedings of the IEEE/CVF international conference on computer vision , pages=

    Emerging properties in self-supervised vision transformers , author=. Proceedings of the IEEE/CVF international conference on computer vision , pages=

  16. [18]

    arXiv preprint arXiv:2508.10104 , year=

    Dinov3 , author=. arXiv preprint arXiv:2508.10104 , year=

    Pith/arXiv arXiv

  17. [19]

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

    Self-supervised learning from images with a joint-embedding predictive architecture , author=. Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition , pages=

  18. [20]

    Proceedings of the IEEE/CVF conference on computer vision and pattern recognition , pages=

    Masked autoencoders are scalable vision learners , author=. Proceedings of the IEEE/CVF conference on computer vision and pattern recognition , pages=

  19. [21]

    5: Improved baselines for agglomerative vision foundation models , author=

    Radiov2. 5: Improved baselines for agglomerative vision foundation models , author=. Proceedings of the Computer Vision and Pattern Recognition Conference , pages=

  20. [22]

    Proceedings of the IEEE/CVF conference on computer vision and pattern recognition , pages=

    Am-radio: Agglomerative vision foundation model reduce all domains into one , author=. Proceedings of the IEEE/CVF conference on computer vision and pattern recognition , pages=

  21. [23]

    arXiv preprint arXiv:2408.00714 , year=

    Sam 2: Segment anything in images and videos , author=. arXiv preprint arXiv:2408.00714 , year=

    Pith/arXiv arXiv

  22. [24]

    arXiv preprint arXiv:2511.16719 , year=

    Sam 3: Segment anything with concepts , author=. arXiv preprint arXiv:2511.16719 , year=

    Pith/arXiv arXiv

  23. [25]

    arXiv preprint arXiv:2511.16624 , year=

    Sam 3d: 3dfy anything in images , author=. arXiv preprint arXiv:2511.16624 , year=

    Pith/arXiv arXiv

  24. [26]

    arXiv preprint arXiv:2212.08751 , year=

    Point-e: A system for generating 3d point clouds from complex prompts , author=. arXiv preprint arXiv:2212.08751 , year=

    Pith/arXiv arXiv

  25. [27]

    arXiv preprint arXiv:2305.02463 , year=

    Shap-e: Generating conditional 3d implicit functions , author=. arXiv preprint arXiv:2305.02463 , year=

    Pith/arXiv arXiv

  26. [28]

    Proceedings of the IEEE/CVF international conference on computer vision , pages=

    Lerf: Language embedded radiance fields , author=. Proceedings of the IEEE/CVF international conference on computer vision , pages=

  27. [29]

    arXiv preprint arXiv:2308.07931 , year=

    Distilled feature fields enable few-shot language-guided manipulation , author=. arXiv preprint arXiv:2308.07931 , year=

    Pith/arXiv arXiv

  28. [30]

    2022 International Conference on 3D Vision (3DV) , pages=

    Neural feature fusion fields: 3d distillation of self-supervised 2d image representations , author=. 2022 International Conference on 3D Vision (3DV) , pages=. 2022 , organization=

    2022

  29. [31]

    Proceedings of the IEEE/CVF conference on computer vision and pattern recognition , pages=

    Clip-nerf: Text-and-image driven manipulation of neural radiance fields , author=. Proceedings of the IEEE/CVF conference on computer vision and pattern recognition , pages=

  30. [32]

    Advances in neural information processing systems , volume=

    Openshape: Scaling up 3d shape representation towards open-world understanding , author=. Advances in neural information processing systems , volume=

  31. [33]

    Proceedings of the IEEE/CVF conference on computer vision and pattern recognition , pages=

    Ulip: Learning a unified representation of language, images, and point clouds for 3d understanding , author=. Proceedings of the IEEE/CVF conference on computer vision and pattern recognition , pages=

  32. [34]

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

    Partfield: Learning 3d feature fields for part segmentation and beyond , author=. Proceedings of the IEEE/CVF International Conference on Computer Vision , pages=

  33. [35]

    Advances in Neural Information Processing Systems , volume=

    Segment anything in 3d with nerfs , author=. Advances in Neural Information Processing Systems , volume=

  34. [36]

    SIGGRAPH Asia 2024 Conference Papers , pages=

    iseg: Interactive 3d segmentation via interactive attention , author=. SIGGRAPH Asia 2024 Conference Papers , pages=

    2024

  35. [37]

    arXiv preprint arXiv:2412.05268 , year=

    Densematcher: Learning 3d semantic correspondence for category-level manipulation from a single demo , author=. arXiv preprint arXiv:2412.05268 , year=

    Pith/arXiv arXiv

  36. [38]

    Liu, Richard and Lang, Itai and Hanocka, Rana , month = jan, year =. Deep. doi:10.48550/arXiv.2601.12527 , abstract =

    work page doi:10.48550/arxiv.2601.12527

  37. [39]

    Proceedings of the IEEE/CVF conference on computer vision and pattern recognition , pages=

    High-resolution image synthesis with latent diffusion models , author=. Proceedings of the IEEE/CVF conference on computer vision and pattern recognition , pages=

  38. [40]

    arXiv preprint arXiv:2010.11929 , year=

    An image is worth 16x16 words: Transformers for image recognition at scale , author=. arXiv preprint arXiv:2010.11929 , year=

    Pith/arXiv arXiv 2010

  39. [41]

    Advances in neural information processing systems , volume=

    Point-voxel cnn for efficient 3d deep learning , author=. Advances in neural information processing systems , volume=

  40. [42]

    arXiv preprint arXiv:2411.07184 , year=

    Sampart3d: Segment any part in 3d objects , author=. arXiv preprint arXiv:2411.07184 , year=

    Pith/arXiv arXiv

  41. [43]

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

    Partslip: Low-shot part segmentation for 3d point clouds via pretrained image-language models , author=. Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition , pages=

  42. [44]

    Proceedings of the International Conference on 3D Vision (3DV) , pages =

    Lang, Itai and Ginzburg, Dvir and Avidan, Shai and Raviv, Dan , title =. Proceedings of the International Conference on 3D Vision (3DV) , pages =

  43. [45]

    ACM SIGGRAPH Asia Conference and Exhibition on Computer Graphics and Interactive Techniques , year =

    PartUV: Part-Based UV Unwrapping of 3D Meshes , author =. ACM SIGGRAPH Asia Conference and Exhibition on Computer Graphics and Interactive Techniques , year =

  44. [46]

    , year =

    Kuhn, Harold W. , year =. Naval Research Logistics Quarterly , volume =

  45. [47]

    2008 , publisher=

    Numerical Geometry of Non-Rigid Shapes , author=. 2008 , publisher=

    2008

  46. [48]

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

    Tong Wu and Jiarui Zhang and Xiao Fu and Yuxin Wang and Jiawei Ren, Liang Pan and Wayne Wu and Lei Yang and Jiaqi Wang and Chen Qian and Dahua Lin and Ziwei Liu , title =. IEEE/CVF Conference on Computer Vision and Pattern Recognition (CVPR) , year=

  47. [49]

    arXiv preprint arXiv:2307.05663 , year=

    Objaverse-XL: A Universe of 10M+ 3D Objects , author=. arXiv preprint arXiv:2307.05663 , year=

    Pith/arXiv arXiv

  48. [50]

    ACM Trans

    Ovsjanikov, Maks and Ben-Chen, Mirela and Solomon, Justin and Butscher, Adrian and Guibas, Leonidas , title =. ACM Trans. Graph. , month = jul, articleno =. 2012 , issue_date =. doi:10.1145/2185520.2185526 , abstract =

    arXiv 2012

  49. [51]

    Blended Intrinsic Maps

    Vladimir Kim and Yaron Lipman and Thomas Funkhouser. Blended Intrinsic Maps. ACM Transactions on Graphics (Proc. SIGGRAPH). 2011

    2011

  50. [52]

    2025 , isbn =

    Uzolas, Lukas and Eisemann, Elmar and Kellnhofer, Petr , title =. 2025 , isbn =. doi:10.1145/3757377.3763974 , booktitle =

    arXiv 2025

  51. [53]

    Subdivision-based Mesh Convolution Networks , journal =

    Shi. Subdivision-based Mesh Convolution Networks , journal =. 2022 , url =

    2022

  52. [54]

    Proceedings of the IEEE international conference on computer vision , pages=

    Multi-view convolutional neural networks for 3d shape recognition , author=. Proceedings of the IEEE international conference on computer vision , pages=

  53. [55]

    Proceedings of the European conference on computer vision (ECCV) , pages=

    3dmv: Joint 3d-multi-view prediction for 3d semantic scene segmentation , author=. Proceedings of the European conference on computer vision (ECCV) , pages=

  54. [56]

    European conference on computer vision , pages=

    Virtual multi-view fusion for 3d semantic segmentation , author=. European conference on computer vision , pages=. 2020 , organization=

    2020

  55. [57]

    Proceedings of the IEEE/CVF international conference on computer vision , pages=

    Clip-fo3d: Learning free open-world 3d scene representations from 2d dense clip , author=. Proceedings of the IEEE/CVF international conference on computer vision , pages=

  56. [58]

    Proceedings of the IEEE/CVF conference on computer vision and pattern recognition , pages=

    Openscene: 3d scene understanding with open vocabularies , author=. Proceedings of the IEEE/CVF conference on computer vision and pattern recognition , pages=

  57. [59]

    arXiv preprint arXiv:2306.13631 , year=

    Openmask3d: Open-vocabulary 3d instance segmentation , author=. arXiv preprint arXiv:2306.13631 , year=

    Pith/arXiv arXiv

  58. [60]

    Advances in neural information processing systems , volume=

    Decomposing nerf for editing via feature field distillation , author=. Advances in neural information processing systems , volume=

  59. [61]

    Proceedings of the IEEE/CVF conference on computer vision and pattern recognition , pages=

    Objaverse: A universe of annotated 3d objects , author=. Proceedings of the IEEE/CVF conference on computer vision and pattern recognition , pages=

  60. [62]

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

    Objaverse++: Curated 3d object dataset with quality annotations , author=. Proceedings of the IEEE/CVF International Conference on Computer Vision , pages=