Pith. sign in

REVIEW 3 major objections 5 minor 52 references

Towards Training-Free Open-World Classification with 3D Generative Models

T0 review · 3 major / 5 minor · reviewed 2026-08-10 · deepseek-v4-flash

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

Pith's one-line read Text-to-3D generative models can supply class prototypes for training-free, pose-invariant open-world 3D classification, beating projection-based methods by wide margins on rotated benchmarks.

desk verdict A plausible new recipe for zero-shot 3D classification using text-to-3D generated anchors, but the headline ModelNet10 result is confounded by retraining TET on the same benchmark, so 'training-free' is overstated. read the letter →

arxiv 2501.17547 v2 pith:5WLBPEC5 submitted 2025-01-29 cs.CV

Xinzhe Xia , Weiguang Zhao , Yuyao Yan , Guanyu Yang , Rui Zhang , Kaizhu Huang , Xi Yang This is my paper · ORCID
classification cs.CV
keywords 3Dopen-worldclassificationtraining-freetext-to-3Dgenerationzero-shotrotationinvariancepointcloudopen-poserecognition
verification ladder T0 review T1 audit T2 compute T3 formal

The pith

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

The reading

This paper claims that a class in the open world—unseen categories, arbitrary orientations, no training data—can be recognized by generating a few anchor point clouds from its text description with a pre-trained text-to-3D model, embedding those clouds with a rotation-invariant encoder, and labeling a test cloud by whichever anchor is most similar under cosine distance. The authors show this training-free recipe on two open-pose benchmarks, reporting overall accuracy gains of 32.7% on ModelNet10‡ and 8.7% on McGill‡ over existing projection-based methods that depend on 2D priors. If the claim holds, 3D generative models store enough shape knowledge to serve as classifiers for novel categories, turning zero-shot 3D classification into a generation-plus-retrieval task that requires no retraining when new classes arrive.

What carries the argument

The load-bearing mechanism is the anchor set: point clouds generated from text descriptions by Shap-E or GaussianDreamer, augmented to match test geometry, and embedded by TET, a rotation-invariant 3D encoder that maps point clouds into a conformal space via steerable spherical neurons, yielding features equivariant to rotation and permutation. These anchors act as class prototypes, and the classifier is nothing more than nearest-prototype cosine similarity, with no learned decision boundary. The authors show that anchor diversity matters: sampling multiple diffusion noises and ensembling anchors from both generators improves per-category accuracy, and the relationship between the number of anchors and accuracy is positive with diminishing returns.

What would settle it

Measure the pairwise feature-space distances between generated anchors and held-out real test samples of the same category on a class set the encoder never saw during pre-training; if same-class anchor-test pairs are not systematically closer than cross-class pairs, the method collapses to chance. A concrete version would apply the pipeline to a held-out subset of ScanObjectNN categories with no ModelNet overlap, where the generative model's idealized shapes would have to bridge to noisy, occluded real scans.

Watch

Extended reading notes

Core claim

The central discovery is that text-to-3D generators produce discriminative prototypes for unseen 3D classes, provided features are extracted by a pose-invariant encoder. The pipeline synthesizes anchor clouds from category names using Shap-E or GaussianDreamer, optionally guided by LLM-written visual descriptions; applies farthest-point sampling, centering, scaling, and random rotation to both anchors and test samples; and maps them through TET, a rotation-invariant point cloud encoder, into a shared space. Classification is simply the nearest-prototype cosine similarity computed in that space. The paper reports that this non-trained scheme beats existing 2D-projection methods by large margins on rotated benchmarks and that the rotation-invariant encoder is indispensable: substituting the rotation-sensitive TAP encoder reduces open-pose accuracy by more than 35 points, while on aligned poses the two encoders perform comparably.

Load-bearing premise

The entire pipeline rests on the empirical assumption that generated anchor clouds and real test clouds of the same category land close together in TET's feature space, so that nearest-prototype cosine similarity is a reliable classifier.

Editorial extensions

If this is right

  • Adding a new category to an open-world classifier costs only the time to generate and embed a handful of anchors, not a retraining run.
  • Rotation invariance in the feature extractor is a prerequisite, not an option: the paper's TET-versus-TAP comparison shows the whole advantage on open-pose data disappears without it.
  • Ensembling anchors from multiple generative models improves per-category accuracy, roughly tracking whichever generator is stronger for a given class, so prototype diversity matters as much as prototype fidelity.
  • The method's ceiling is set by the text-to-3D generator: for categories with low accuracy, richer LLM descriptions can help or hurt depending on how faithfully the description captures the category's core visual features.

Reading between the lines

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

  • Editorial inference: the generation-plus-retrieval recipe is modular and could transfer to other modalities, such as generating audio or video prototypes from text for zero-shot classification, wherever a rotation- or transformation-invariant embedding exists.
  • Editorial inference: the sharp drop in accuracy when the encoder is pre-trained on a different distribution (Table 5: from 90.2% to 54.6% on ModelNet10‡) suggests that pairing TET with a representation space explicitly aligned between generated and real scans is a direct next step toward harder open-world settings.
  • Editorial inference: because anchors are generated per category name, the method implicitly models category typicality rather than instance diversity; a testable fix would be to condition multiple anchors on visual subtypes of a category, which could be discovered by clustering the first-pass predictions.
Share X Bluesky LinkedIn Reddit HN

Signed reviews

No signed human review yet.

Editorial analysis

A structured set of objections, weighed in public.

Desk editor's note, referee report, and a circularity audit.

Referee Report

3 major / 5 minor

Summary. The paper proposes a training-free pipeline for 3D open-world classification that uses pre-trained text-to-3D generative models (Shap-E and GaussianDreamer) to synthesize anchor point clouds for novel category names, optionally enriched by LLM-generated descriptions. Anchors and test samples are processed by a frozen rotation-invariant 3D encoder (TET), and classification is performed by nearest-prototype cosine similarity in the feature space. Experiments on the open-pose benchmarks ModelNet10‡ and McGill‡ report state-of-the-art overall accuracy (59.0% and 47.8%, respectively), with ablations on the number of anchors, choice of generative model, choice of representation model, and encoder pretraining source.

Significance. If the central claim is cleanly established, the paper would introduce a genuinely new mechanism for open-world 3D classification: using 3D generative priors to synthesize class prototypes, thereby avoiding 2D projection and task-specific retraining. The paper has several strengths: it is honest about the negative effect of LLM-generated prompts, it includes a useful comparison between rotation-variant (TAP) and rotation-invariant (TET) encoders, and it quantifies the impact of the encoder's pretraining source in Table 5. However, the current experimental protocol does not cleanly support the 'training-free' and 'open-world' claims, because the ModelNet10 result is obtained after training a new representation model on a curated subset of the same benchmark family, and key hyperparameters are selected on the test sets. These issues are load-bearing for the paper's main assertions.

major comments (3)
  1. [§4.2, Table 5] The 'training-free' claim is contradicted by the experimental protocol. For ModelNet10‡, the authors state: 'we manually exclude the overlapping categories from ModelNet40 to form ModelNet30, and subsequently trained a new representation model on this adjusted dataset for the ModelNet10 tests.' This is task-specific training on a dataset derived from the same benchmark family as the test set. Table 5 quantifies the effect: TET pretrained on full ModelNet40 (10 overlapping categories) reaches 90.2% oAcc, but on non-overlapping ModelNet30 it drops to 59.0%, and on ScanObjectNN it is 54.6%. Thus the reported 59.0% on ModelNet10‡ cannot be attributed to the generative-anchor mechanism alone; the encoder's pretraining distribution is a dominant factor. A training-free open-world claim requires a frozen encoder whose pretraining data is disjoint from the test categories and domain, or at minimum a clear statement that 'training-free' refers only to the classifier head and not to the representation model.
  2. [§4.3, Fig. 3, Tables 2–3] The final configuration is selected on the same test sets used for reporting. The number of anchors (Na=7), the ensemble of Shap-E and GaussianDreamer, the choice of TET over TAP, and the exclusion of LLM-generated prompts in Table 1 are all decided after observing test-set accuracy. No validation split is used, and no variance estimates are provided across runs or generator seeds. Consequently, the reported improvements over baselines may reflect selection bias rather than a robust property of the pipeline. Please report mean±standard deviation over multiple runs and/or use a disjoint validation set for hyperparameter choices, or demonstrate explicitly that the conclusions are insensitive to these choices.
  3. [§4.4, Table 7] LLM-generated descriptions improve overall accuracy on ModelNet10‡ (oAcc 59.5 vs. 57.8, mAcc 60.8 vs. 59.2 with Shap-E), yet these results are excluded from the main comparison in Table 1 because per-category performance is mixed. This is a post hoc selection on the test data. Moreover, Fig. 2 and Section 3.3 present LLM-enriched descriptions as part of the proposed pipeline, so excluding them from the headline results without a principled criterion is inconsistent. Either include LLM prompts with a fixed, pre-specified prompt-selection rule, or clearly present LLM prompting as an optional component and evaluate it on a validation set rather than the test set.
minor comments (5)
  1. [Abstract] The abstract states a '32.0%' overall accuracy improvement on ModelNet10, while the main text and Table 1 report '32.7%'. This inconsistency should be corrected.
  2. [Throughout] There are several typographical errors, e.g., 'leverageing' in the abstract, 'donates' for 'denotes' in Equations (3) and (5), and 'donates' elsewhere. A careful proofread is needed.
  3. [Eq. (4) and §4.1.3] Equation (4) defines the anchor feature as an element of R^{N_p × D}, but the implementation text says TET outputs (256,1) and TAP outputs (256,384) pooled to (256,1). Please clarify whether the extracted feature is a per-point tensor or a global descriptor, and make the notation consistent.
  4. [Table 3] The header 'RMs' in Table 3 is misleading; the table compares generative models (Shap-e, GD, Ensemble), not representation models. Change the header to 'Generative Model'.
  5. [Fig. 1] The label 'StyleOrientation' in Figure 1 appears to be a typo or artifact; also the figure caption could more clearly distinguish the proposed pipeline from prior 2D-projection methods.

Circularity Check

0 steps flagged · score 2.0 of 10

No significant circularity: classification is a frozen nearest-prototype rule with external anchor generation, so the headline accuracies are measurements rather than fitted predictions; the minor score reflects the authors' own OP3D benchmark self-citation and a confounded auxiliary evidence claim, neither of which drives the main numbers by construction.

full rationale

The derivation chain is empirical and open: Eq. (2) generates anchors from category names with external text-to-3D models (Shap-E, GaussianDreamer); Eqs. (4) and (6) extract features with a frozen TET encoder; Eqs. (7)-(8) apply cosine-similarity nearest-prototype classification. No classifier parameter is fitted to the test labels, and the headline ModelNet10 result (59.0% oAcc) uses TET pre-trained on ModelNet30 with the ten test categories excluded, while the McGill result uses TET pre-trained on ModelNet40 with no category overlap. Thus the reported accuracies are not equivalent to the method's inputs by construction. The paper does cite its own OP3D benchmark [46] and prior training-free grasping work [44], but these are evaluation infrastructure and related work, not load-bearing derivation. Two caveats are worth noting but are not circularity: Section 4.3.4 uses the overlapping ModelNet40-pretrained TET's 90.2% accuracy to 'demonstrate' that generated anchors are informative, which is confounded by the encoder's supervision on those same categories; and Section 4.2 retrains TET on ModelNet30 despite the 'training-free' label. These weaken the strength of the claims but do not make the central non-overlap results reduce to a fit or a self-citation chain.

Assumptions & free parameters 1 free parameters · 4 assumptions · 0 invented entities

The pipeline rests on two families of external, unverified bridges: text-to-3D generators produce category-representative geometry, and a pre-trained encoder embeds both generated and real point clouds into a shared rotation-invariant space. The paper provides empirical evidence for both on two benchmarks but no guarantees. The only numerical quantity chosen by hand is the anchor count Na=7.

free parameters (1)
  • Number of anchor samples Na = 7
    Chosen from Fig. 3 as the point of diminishing returns on the same test benchmarks used for the headline results (Section 4.3.1).
assumptions (4)
  • domain assumption Text-to-3D generators (Shap-E, GaussianDreamer) produce point clouds whose category-level geometry matches real objects well enough for recognition.
    Invoked in Eq. 2 and Section 3.2; if the anchors are off-distribution, cosine similarity in Eq. 7 cannot classify.
  • domain assumption TET features are rotation-invariant and form a semantically organized metric space for point clouds.
    Relied on in Sections 3.3 and 3.4 and tested in Section 4.3.3; invariance is claimed for the encoder, not proven for generated anchors.
  • domain assumption Farthest point sampling, origin shifting, scaling and random rotation preserve the category label of anchor and test point clouds.
    Used in Eq. 3 and Eq. 5; if augmentation changed semantics, the prototypes would be corrupted.
  • domain assumption Cosine similarity between pooled point-level features is a valid classification rule.
    The entire inference phase (Eqs. 7-8) assumes nearest-prototype matching in TET space separates categories.

how reviews work

0 comments
Cite this review

Pith. "Pith review of Towards Training-Free Open-World Classification with 3D Generative Models." pith.science (2026). https://pith.science/paper/5WLBPEC5

@misc{pith2026250117547,
  author       = {Pith},
  title        = {Pith review of: Towards Training-Free Open-World Classification with 3D Generative Models},
  year         = {2026},
  howpublished = {\url{https://pith.science/paper/5WLBPEC5}},
  note         = {Machine review of arXiv:2501.17547}
}
read the original abstract

3D open-world classification is a challenging yet essential task in dynamic and unstructured real-world scenarios, requiring both open-category and open-pose recognition. To address these challenges, recent wisdom often takes sophisticated 2D pre-trained models to provide enriched and stable representations. However, these methods largely rely on how 3D objects can be projected into 2D space, which is unfortunately not well solved, and thus significantly limits their performance. Unlike these present efforts, in this paper we make a pioneering exploration of 3D generative models for 3D open-world classification. Drawing on abundant prior knowledge from 3D generative models, we additionally craft a rotation-invariant feature extractor. This innovative synergy endows our pipeline with the advantages of being training-free, open-category, and pose-invariant, thus well suited to 3D open-world classification. Extensive experiments on benchmark datasets demonstrate the potential of generative models in 3D open-world classification, achieving state-of-the-art performance on ModelNet10 and McGill with 32.0% and 8.7% overall accuracy improvement, respectively.

Figures

Figures reproduced from arXiv: 2501.17547 by the authors.

Figure 1
Figure 1. Comparison of Pipelines: (a) SOTA methods [46, 47] project 3D samples into 2D images to harness 2D prior knowledge for novel category recognition but are sensitive to pose changes, (b) The proposed pipeline applies 3D prior knowledge to generate anchor samples for novel categories, embedding them in a rotation-invariant space for effective performance in open-pose scenarios. Predictions are com￾puted via feature sim… view at source ↗
Figure 2
Figure 2. Overview of the network architecture with two components. (1) Open-world classifier: Category names are input into ChatGPT to create descriptions, utilized by a pre-trained text-to-3D model to generate anchor samples. These samples are augmented to align with test samples. A pre-trained 3D model extracts features, marking them as prototypes for classification. (2) Test sample inference: Test samples are augmented an… view at source ↗
Figure 3
Figure 3. Comparison on the Number of Anchor Samples 𝑁𝑎. 4.2 Performance Comparison Our approach is compared against six SOTA methods: Point￾CLIP [42], ULIP [38], ReconCLIP [21], CLIP2Point [8], Point￾CLIPv2 [47], and OP3D [46], on the open-pose classifica￾tion datasets ModelNet10‡ and McGill‡ . To ensure fairness, we utilize the pre-trained models from the official GitHub repositories provided by the respective papers (TAP [… view at source ↗
Figures from the paper (2 more)
Figure 4
Figure 4. Figure 4: t-SNE visualizations of feature representations extracted by TAP and TET under aligned-pose and open￾pose settings on the ModelNet10 dataset. Different colors indicate different object categories according to the ground truth labels. Triangles represent generated ancho…
Figure 5
Figure 5. Figure 5: Visualization comparison of "dresser" point clouds from original category names (CN) and ChatGPT-generated prompts (CD), which enhance anchors’ diversity. indicating that out-of-distribution (OOD) issues persist in rotation-invariant models. On the other hand, the resu…

Discussion (0). Continue with ORCID to comment.

Reference graph

Works this paper leans on

52 extracted references · 46 canonical work pages

  1. [1]

    Ali Cheraghian, Shafin Rahman, Dylan Campbell, and Lars Petersson

    work page

  2. [2]

    Ali Cheraghian, Shafin Rahman, Townim F Chowdhury, Dylan Camp- bell, and Lars Petersson. 2022. Zero-shot learning on 3d point cloud objects and beyond. IJCV 130, 10 (2022), 2364–2384

    work page 2022

  3. [3]

    Christopher Choy, JunYoung Gwak, and Silvio Savarese. 2019. 4d spatio-temporal convnets: Minkowski convolutional neural networks. In CVPR. 3075–3084

    work page 2019

  4. [4]

    Yun Hao, Yukun Su, Guosheng Lin, Hanjing Su, and Qingyao Wu. 2023. Contrastive Generative Network with Recursive-Loop for 3D point cloud generalized zero-shot classification. PR 144 (2023), 109843

    work page 2023

  5. [5]

    Deepti Hegde, Jeya Maria Jose Valanarasu, and Vishal Patel. 2023. Clip goes 3d: Leveraging prompt tuning for language grounded 3d recognition. In ICCV. 2028–2038

    work page 2023

  6. [6]

    Jonathan Ho, Ajay Jain, and Pieter Abbeel. 2020. Denoising Diffusion Probabilistic Models. In NeurIPS, Vol. 33. 6840–6851

    2020

  7. [7]

    Wenbo Hu, Hengshuang Zhao, Li Jiang, Jiaya Jia, and Tien-Tsin Wong

    work page

  8. [8]

    Tianyu Huang, Bowen Dong, Yunhan Yang, Xiaoshui Huang, Ryn- son WH Lau, Wanli Ouyang, and Wangmeng Zuo. 2023. Clip2point: Transfer clip to point cloud classification with image-depth pre- training. In ICCV. 22157–22167

    work page 2023

Show all 52 references
  1. [9]

    Chenru Jiang, Wuwei Ma, Kaizhu Huang, Qiufeng Wang, Xi Yang, Weiguang Zhao, Junwei Wu, Xinheng Wang, Jimin Xiao, and Zhenxing Niu. 2025. Revisiting 3D point cloud analysis with Markov process. PR 158 (2025), 110997

    2025

  2. [10]

    Heewoo Jun and Alex Nichol. 2023. Shap-e: Generating conditional 3d implicit functions. arXiv (2023)

    2023

  3. [11]

    Minghua Liu, Ruoxi Shi, Kaiming Kuang, Yinhao Zhu, Xuanlin Li, Shizhong Han, Hong Cai, Fatih Porikli, and Hao Su. 2023. OpenShape: Scaling Up 3D Shape Representation Towards Open-World Under- standing. In NeurIPS, A. Oh, T. Naumann, A. Globerson, K. Saenko, M. Hardt, and S. Le...

    2023

  4. [12]

    Shitong Luo and Wei Hu. 2021. Diffusion probabilistic models for 3d point cloud generation. In CVPR. 2837–2845

    2021

  5. [13]

    Daniel Maturana and Sebastian Scherer. 2015. Voxnet: A 3d convolu- tional neural network for real-time object recognition. 922–928

    2015

  6. [14]

    Soroush Mehraban, Vida Adeli, and Babak Taati. 2024. Motionag- former: Enhancing 3d human pose estimation with a transformer- gcnformer network. In W ACV. 6920–6930

    2024

  7. [15]

    Pavlo Melnyk, Andreas Robinson, Michael Felsberg, and Mårten Wadenbäck. 2024. TetraSphere: A Neural Descriptor for O (3)-Invariant Point Cloud Analysis. In CVPR. 5620–5630

    2024

  8. [16]

    Tomas Mikolov, Ilya Sutskever, Kai Chen, Greg S Corrado, and Jeff Dean. 2013. Distributed representations of words and phrases and their compositionality. NeurIPS 26 (2013)

    2013

  9. [17]

    Alex Nichol, Prafulla Dhariwal, Aditya Ramesh, Pranav Shyam, Pamela Mishkin, Bob McGrew, Ilya Sutskever, and Mark Chen. 2021. GLIDE: Towards Photorealistic Image Generation and Editing with Text- Guided Diffusion Models. arXiv abs/2112.10741 (2021)

    2021 arXiv

  10. [18]

    Alex Nichol, Heewoo Jun, Prafulla Dhariwal, Pamela Mishkin, and Mark Chen. 2022. Point-e: A system for generating 3d point clouds from complex prompts. arXiv (2022)

    2022

  11. [19]

    Jeffrey Pennington, Richard Socher, and Christopher D Manning. 2014. Glove: Global vectors for word representation. In EMNLP. 1532–1543

    2014

  12. [20]

    Charles R Qi, Hao Su, Kaichun Mo, and Leonidas J Guibas. 2017. Point- net: Deep learning on point sets for 3d classification and segmentation. In CVPR. 652–660

    2017

  13. [21]

    Zekun Qi, Runpei Dong, Guofan Fan, Zheng Ge, Xiangyu Zhang, Kaisheng Ma, and Li Yi. 2023. Contrast with reconstruct: Contrastive 3d representation learning guided by generative pretraining. In ICML. 28223–28243

    2023

  14. [22]

    Alec Radford, Jong Wook Kim, Chris Hallacy, Aditya Ramesh, Gabriel Goh, Sandhini Agarwal, Girish Sastry, Amanda Askell, Pamela Mishkin, Jack Clark, et al. 2021. Learning transferable visual models from natural language supervision. In ICML. 8748–8763

    2021

  15. [23]

    Blaine Rister, Mark A Horowitz, and Daniel L Rubin. 2017. Volumetric image registration from invariant keypoints. IEEE TIP 26, 10 (2017), 4900–4910

    2017

  16. [24]

    Robin Rombach, Andreas Blattmann, Dominik Lorenz, Patrick Esser, and Björn Ommer. 2022. High-resolution image synthesis with latent diffusion models. In CVPR. 10684–10695

    2022

  17. [25]

    Sitian Shen, Zilin Zhu, Linqian Fan, Harry Zhang, and Xinxiao Wu

  18. [26]

    Kaleem Siddiqi, Juan Zhang, Diego Macrini, Ali Shokoufandeh, Sylvain Bouix, and Sven Dickinson. 2008. Retrieving articulated 3-D models using medial surfaces. Machine Vision and Applications 19 (2008), 261–275

    2008

  19. [27]

    Jascha Sohl-Dickstein, Eric Weiss, Niru Maheswaranathan, and Surya Ganguli. 2015. Deep unsupervised learning using nonequilibrium thermodynamics. In ICML. 2256–2265

    2015

  20. [28]

    Mikaela Angelina Uy, Quang-Hieu Pham, Binh-Son Hua, Duc Thanh Nguyen, and Sai-Kit Yeung. 2019. Revisiting Point Cloud Classification: A New Benchmark Dataset and Classification Model on Real-World Data. In International Conference on Computer Vision (ICCV)

    2019

  21. [29]

    Gomez, Lukasz Kaiser, and Illia Polosukhin

    Ashish Vaswani, Noam Shazeer, Niki Parmar, Jakob Uszkoreit, Llion Jones, Aidan N. Gomez, Lukasz Kaiser, and Illia Polosukhin. 2017. Attention is All you Need. In NeurIPS. 5998–6008

    2017

  22. [30]

    Peng-Shuai Wang. 2023. Octformer: Octree-based transformers for 3d point clouds. ACM TOG 42, 4 (2023), 1–11

    2023

  23. [31]

    Peng-Shuai Wang. 2023. Octformer: Octree-based transformers for 3d point clouds. TOG 42, 4 (2023), 1–11

    2023

  24. [32]

    Peng-Shuai Wang, Yang Liu, Yu-Xiao Guo, Chun-Yu Sun, and Xin Tong

  25. [33]

    Ziyi Wang, Xumin Yu, Yongming Rao, Jie Zhou, and Jiwen Lu. 2023. Take-a-photo: 3d-to-2d generative pre-training of point cloud models. In ICCV. 5640–5650

    2023

  26. [34]

    Xiaoyang Wu, Li Jiang, Peng-Shuai Wang, Zhijian Liu, Xihui Liu, Yu Qiao, Wanli Ouyang, Tong He, and Hengshuang Zhao. 2024. Point Transformer V3: Simpler Faster Stronger. In CVPR. 4840–4851

    2024

  27. [35]

    Zhirong Wu, Shuran Song, Aditya Khosla, Fisher Yu, Linguang Zhang, Xiaoou Tang, and Jianxiong Xiao. 2015. 3d shapenets: A deep repre- sentation for volumetric shapes. In CVPR. 1912–1920. Towards Training-Free Open-World Classification with 3D Generative Models Conference’17, J...

    2015

  28. [36]

    Zijie Wu, Yaonan Wang, Mingtao Feng, He Xie, and Ajmal Mian. 2023. Sketch and Text Guided Diffusion Model for Colored Point Cloud Generation. In ICCV. 8929–8939

    2023

  29. [37]

    Runsen Xu, Xiaolong Wang, Tai Wang, Yilun Chen, Jiangmiao Pang, and Dahua Lin. 2024. Pointllm: Empowering large language models to understand point clouds. In ECCV. Springer, 131–147

    2024

  30. [38]

    Le Xue, Mingfei Gao, Chen Xing, Roberto Martín-Martín, Jiajun Wu, Caiming Xiong, Ran Xu, Juan Carlos Niebles, and Silvio Savarese. 2023. ULIP: Learning a unified representation of language, images, and point clouds for 3D understanding. In CVPR. 1179–1189

    2023

  31. [39]

    Le Xue, Ning Yu, Shu Zhang, Artemis Panagopoulou, Junnan Li, Roberto Martín-Martín, Jiajun Wu, Caiming Xiong, Ran Xu, Juan Car- los Niebles, et al . 2024. Ulip-2: Towards scalable multimodal pre- training for 3d understanding. In CVPR. 27091–27101

    2024

  32. [40]

    Cheng-Kun Yang, Min-Hung Chen, Yung-Yu Chuang, and Yen-Yu Lin

  33. [41]

    Taoran Yi, Jiemin Fang, Junjie Wang, Guanjun Wu, Lingxi Xie, Xi- aopeng Zhang, Wenyu Liu, Qi Tian, and Xinggang Wang. 2024. Gaus- sianDreamer: Fast Generation from Text to 3D Gaussians by Bridging 2D and 3D Diffusion Models. In CVPR. 6796–6807

    2024

  34. [42]

    Junbo Zhang, Runpei Dong, and Kaisheng Ma. 2023. Clip-fo3d: Learn- ing free open-world 3d scene representations from 2d dense clip. In ICCV. 2048–2059

    2023

  35. [43]

    Hengshuang Zhao, Li Jiang, Jiaya Jia, Philip HS Torr, and Vladlen Koltun. 2021. Point transformer. In ICCV. 16259–16268

    2021

  36. [44]

    Weiguang Zhao, Chenru Jiang, Chengrui Zhang, Jie Sun, Yuyao Yan, Rui Zhang, and Kaizhu Huang. 2024. Towards Cross-device and Training-free Robotic Grasping in 3D Open World. arXiv preprint arXiv:2411.18133 (2024)

    2024 arXiv

  37. [45]

    Weiguang Zhao, Yuyao Yan, Chaolong Yang, Jianan Ye, Xi Yang, and Kaizhu Huang. 2023. Divide and conquer: 3d point cloud instance segmentation with point-wise binarization. In ICCV. 562–571

    2023

  38. [46]

    Weiguang Zhao, Guanyu Yang, Rui Zhang, Chenru Jiang, Chaolong Yang, Yuyao Yan, Amir Hussain, and Kaizhu Huang. 2025. Open-Pose 3D zero-shot learning: Benchmark and challenges. NN 181 (2025), 106775

    2025

  39. [47]

    Xiangyang Zhu, Renrui Zhang, Bowei He, Ziyao Zeng, Shanghang Zhang, and Peng Gao. 2023. PointCLIP V2: Adapting CLIP for Powerful 3D Open-world Learning. In ICCV. 2639–2650

    2023

  40. [2017]

    TOG 36, 4 (2017), 1–11

    O-cnn: Octree-based convolutional neural networks for 3d shape analysis. TOG 36, 4 (2017), 1–11

    2017

  41. [2020]

    In W ACV

    Transductive zero-shot learning for 3d point cloud classification. In W ACV. 923–933

  42. [2021]

    Bidirectional projection network for cross dimension scene understanding. In CVPR. 14373–14382

  43. [2023]

    2D-3D interlaced transformer for point cloud segmentation with scene-level supervision. In ICCV. 977–987

  44. [2024]

    In W ACV

    DiffCLIP: Leveraging Stable Diffusion for Language Grounded 3D Classification. In W ACV. 3596–3605

Pith tools

Reviewed August 10, 2026 · model on record in the stance chip above.