Pith. sign in

REVIEW 3 major objections 9 minor 63 references

ThinkAfford: Affordance-Centric Reasoning for Fine-Grained 3D Grounding in Cluttered Scenes

T0 review · 3 major / 9 minor · reviewed 2026-08-12 · deepseek-v4-flash

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

Pith's one-line read ThinkAfford claims that fine-grained 3D affordance grounding improves when proposal generation and instruction reasoning are decoupled, reaching 25.46 AP25 and 10.69 AP50 on the official SceneFun3D evaluator.

desk verdict A sensible two-stage decomposition with believable internal ablations, but the SOTA claim rests on unreleased baseline reproductions—worth refereeing, not desk rejecting. read the letter →

arxiv 2608.10981 v1 pith:4WPO7KBJ submitted 2026-08-11 cs.CV

Xinrui Lin , Sha Zhang , Shumin Wang , Zenghuan Zhu , Jiajun Deng , Yanyong Zhang This is my paper · ORCID
classification cs.CV
keywords 3Daffordancegroundingtask-drivenproposalgenerationvisualreasoninggrouprelativepolicyoptimizationvision-languagemodelsfine-grainedclutteredscenes
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

The paper is trying to show that fine-grained 3D affordance grounding fails for two separable reasons, and that addressing them separately is better than addressing either alone. The target in this task is a small interaction region, such as a drawer handle or a socket, so a proposal that returns the whole object is wrong even if it is semantically correct. ThinkAfford splits the pipeline: the Affordance Proposal Generation module deliberately over-produces fine-grained candidates from interaction-conditioned heatmaps, and the Visual-Prompted Affordance Reasoning module selects identifiers from a labeled overlay using the full instruction. A Group Relative Policy Optimization (GRPO) objective rewards selections by their lifted 3D overlap with ground truth, teaching the reasoner to handle spatial, ordinal, and part-whole relations. On the official SceneFun3D evaluation the full system reaches 25.46 AP25 and 10.69 AP50, and module-level diagnostics report 77.5% proposal coverage at IoU 0.25 and 72.1% selection accuracy on covered queries.

What carries the argument

The load-bearing mechanism is the labeled proposal overlay together with a discrete selection action. APG turns each affordance type into a heatmap, thresholds it, and clusters connected components into a variable-length set of candidates, each tagged with a letter identifier, so the reasoning module operates on a finite list of physically distinct regions rather than generating open-ended masks. The second carrying mechanism is the GRPO reward: every sampled think-then-answer response is parsed into identifiers, each proposal is lifted to 3D with a visibility-aware z-buffer and local-plane disambiguation, and point-level IoU at threshold $\rho=0.25$ defines which proposals count as covering the target; group-relative advantages then let the policy learn relational distinctions from the differences among similar candidates.

What would settle it

Take a random stratified sample of queries per affordance type, manually inspect the lifted APG proposals against the ground-truth contact regions, and rerun the official evaluator with the positive IoU threshold raised from 0.25 to 0.4; a sharp coverage drop or frequent visible misalignment of thin parts would weaken the central claim that APG generates high-recall interaction-level proposals.

Watch

Extended reading notes

Core claim

ThinkAfford's central claim is that task-driven 3D grounding should be a two-stage process with distinct measures: first generate a high-recall set of interaction-level proposals, then reason over those proposals to select the correct ones. The paper argues that this decomposition isolates the two failure modes that plague existing systems, missing the interaction region and confusing visually similar alternatives, and that each stage can then be optimized with an appropriate supervision. APG uses learnable affordance prompts with multi-level visual features to predict heatmaps and extract a variable number of small candidates without needing parsed object or part names; VPAR consumes labeled overlays and returns identifiers in a think-then-answer format, trained with GRPO where proposal-level rewards come from lifting each candidate to 3D and comparing point-level IoU with ground truth. The reported numbers support the claim that both stages matter: replacing APG with parsed-object segmentation costs 7.16 AP25, and removing GRPO costs 8.66 AP25.

Load-bearing premise

Everything depends on the assumption that lifting 2D proposal masks into the 3D point cloud faithfully matches the true visible interaction region; if the z-buffer and local-plane disambiguation bias that mapping, the training rewards and all reported coverage and selection numbers shift.

Editorial extensions

If this is right

  • Systems should track proposal coverage and selection accuracy separately; an end-to-end score alone cannot show whether errors come from missed candidates or wrong choices.
  • Interaction-conditioned heatmaps can replace parsed object or part names as the prompt for segmenting functional regions, avoiding the granularity mismatch that produces whole-object masks.
  • GRPO with lifted 3D overlap rewards is a viable way to teach a vision-language model relational instruction following without hand-built negatives.
  • The smaller gain at stricter IoU thresholds indicates that once the right region is found, boundary tightness is the remaining bottleneck.
  • The complementary residual errors, 22.5% proposals missed at IoU 0.25 and 27.9% misselection on covered queries, imply that further progress needs both better proposal tightness and better relation-aware selection.

Reading between the lines

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

  • Editorial inference: the same search-then-select decomposition could generalize to other fine-grained localization tasks with many small repeated parts, such as part-level picking in assembly or identifying interaction points in surgical scenes, wherever a high-recall candidate stage and a language-guided selection stage can be separated.
  • Editorial inference: a stress test with the positive threshold raised from $\rho=0.25$ to, say, 0.4 would reveal how much of the reported advantage is robust target identification rather than loose-overlap tolerance; the paper does not report this sweep.
  • Editorial inference: the qualitative robot demonstrations set up a measurable downstream claim, that higher selection accuracy should raise first-attempt manipulation success when the localized contact point is fed to a controller, but the paper does not evaluate that success rate.
Share X Bluesky LinkedIn Reddit HN

Editorial analysis

A structured set of objections, weighed in public.

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

Referee Report

3 major / 9 minor

Summary. ThinkAfford proposes a two-module decomposition for task-driven 3D affordance grounding on SceneFun3D: the Affordance Proposal Generation (APG) module produces variable-length, interaction-level 2D proposals from affordance-conditioned heatmaps, trained with visibility-aware projections of 3D ground-truth masks, and the Visual-Prompted Affordance Reasoning (VPAR) module, a VLM fine-tuned with GRPO using proposal-level rewards computed from lifted 3D overlap, selects proposal identifiers in a think-then-answer format; the selected masks are lifted and fused by weighted voting into a final 3D mask. On the official SceneFun3D validation evaluator the method reports 25.46 AP25 and 10.69 AP50, above the compared baselines, with module-level diagnostics of 77.5% R@25 proposal coverage and 72.1% selection accuracy on APG-covered queries. The paper additionally reports zero-shot transfer to four self-annotated cross-dataset sets and a qualitative real-robot deployment.

Significance. If the results hold, the paper's contribution is a clean and separately diagnosable separation of proposal coverage from instruction-grounded selection, plus a GRPO training scheme whose reward is anchored in lifted 3D overlap rather than in parsed object names. The ablations are internally consistent, with the largest drops when removing GRPO and APG, and the main evaluation uses the official SceneFun3D evaluator; I also credit the authors' candor, since Sec. 4.3 explicitly calls the qualitative examples illustrative rather than evidence of solved boundary errors, Sec. 4.4 states that coverage and selection numbers are not numerically composable, and Sec. 4.6 disclaims the deployment as not a manipulation success-rate study. The main risks are verification gaps rather than circularity: four baselines in Table 1 are unreleased self-reproductions, the GRPO reward shares its positive threshold with the definition of the APG-covered subset, and no variance estimates accompany small ablation deltas. These issues are fixable within the manuscript's scope, which leads me to major revision rather than rejection.

major comments (3)
  1. [Sec. 4.1, Table 1] The headline claim of state-of-the-art results rests on four self-reproductions (Mask3D-F, TASA, AffordBot, Fun3DU), but the manuscript does not state for any of these baselines which VLM or segmentor was used, how many views were processed, or which prompt templates were employed, so the reported margins (+12.90 AP25 over Fun3DU, +4.32 AP50 over AffordBot) cannot be checked by readers, and the possibility that weaker components were used in those runs is a concrete correctness risk for the comparative claim. Please release the full reproduction configuration (model versions, view budgets, prompts, conversion code) and, as a minimal check, re-run Fun3DU and AffordBot with a VLM of capability comparable to the one used in ThinkAfford's VPAR. In addition, the related work cites OpenMask3D-F, the SceneFun3D benchmark's own functional-3D baseline, but Table 1 omits it; include it in the table or give an explicit reason for its exclusion.
  2. [Sec. 3.3, Eq. (2); Table 4] The positive threshold ρ=0.25 plays a double role: it defines the GRPO reward target set C⋆v (Sec. 3.3: 'Proposals with ov,j ≥ ρ form the target set') and it defines the APG-covered subset on which Table 4 measures selection accuracy, since samples whose best proposal fails to reach ρ are excluded from VPAR training and from the selection-accuracy denominator. This coupling means that the 72.1% selection accuracy and the GRPO-vs-SFT gains are established only under a coarse IoU criterion where proposals that would fail AP50 still count as positive; the paper's own observation that the averaged-AP margin over Fun3DU is only +0.52 (Table 1) is consistent with weak high-IoU alignment, so the claim that GRPO aligns VPAR selection with final 3D grounding needs support from a sensitivity analysis at ρ=0.5 or with a graded IoU reward, together with end-to-end AP reported separately on the covered and non-covered subsets.
  3. [Sec. 4.4, Tables 2 and 4] All reported numbers come from a single training run of APG and a single GRPO run of VPAR; several component claims rest on small deltas (GRPO without format reward −0.56 AP25, uniform view weights −1.26 AP25, APG without visibility-aware projection −1.66 AP25), and the text describes these as 'consistent gains' without any variance estimate, so signal and noise cannot be separated at these magnitudes, even though the large main-effect deltas (removing GRPO, −8.66 AP25; replacing APG with parsed-object Grounded-SAM, −7.16 AP25) are likely robust. Please run at least three seeds for the full model and for the two largest ablation variants, and report mean ± std or a paired test for the Table 2 deltas.
minor comments (9)
  1. [Sec. 3.2] The heatmap threshold γ and the DBSCAN eps and min_samples values used in Proposal Extraction are never reported, which prevents reproduction of the variable-length proposal sets that underlie all coverage and selection numbers.
  2. [Sec. 3.3, Eq. (2)] λfmt is described only as 'small'; please report its exact value and the exact format-reward definition, since Eq. (2) and the corresponding ablation in Table 2 depend on it.
  3. [Sec. 3.1 vs. Sec. 4.1] It is unclear whether K=25 is the number of views retained after the visibility and relation pruning or the uniform sampling interval; clarify the relationship between the pruning stage and K.
  4. [Sec. 4.5, Table 5] The 20-scene, 15-instruction cross-dataset set has no description of the annotation protocol, no inter-annotator agreement statistics, and no release plan; without these, the transfer claim is not independently checkable.
  5. [Table 1 caption] Please state why the AP column is '–' for OpenMask3D and LERF and clarify that AP is the 0.50:0.95 average, so that the table is self-contained for readers.
  6. [Sec. 3.2] APG is trained with the ground-truth affordance type a but used at inference with the VLM-parsed type a⋆, and the manuscript never measures how often a⋆ is correct; report this parsing accuracy, since parsing errors propagate directly into the APG heatmaps.
  7. [Figure 1] The figure contains a stray text fragment ('画一段video') that appears to be a leftover annotation; remove it or translate it into the paper's language.
  8. [Sec. 4.2] The sentence about overlapping relation subsets appears immediately after Table 1, but Table 1 contains no relation slices; move this sentence to the discussion of Tables 3 and 4.
  9. [References] Several typographical errors remain: 'Yeshwanthetal.' is missing a space, the Radford reference has a broken journal rendering ('PmL R'), and 'laser_scan_5mmsupport' appears in the Sec. 4.4 prose.

Circularity Check

0 steps flagged · score 0.0 of 10

No circular derivation: the method trains on external 3D ground truth and is evaluated on a held-out split; module diagnostics are honestly conditioned rather than definitionally forced.

full rationale

The claimed derivation is self-contained against external supervision. APG is trained with pixel-wise BCE on 2D targets Y_a^v obtained by projecting ground-truth 3D instance masks through a visibility-aware z-buffer (Sec. 3.2, Eq. 1), and VPAR is optimized with GRPO rewards computed from point-level IoU between lifted proposals and the ground-truth mask M^3D_gt (Sec. 3.3, Eqs. 2-5). Neither training target is defined in terms of the model's own output, and all headline numbers in Table 1 are reported on the held-out 30-scene validation split under the official evaluator (Sec. 4.1). No load-bearing self-citation appears: the architecture cites OOAL, CoOp, DINOv2, CLIP, Qwen-VL, and GRPO, all external; no uniqueness theorem or prior result by the present authors is invoked to force the design. The only mild conditioning caveat is that the GRPO target set C*_v and the 'APG-covered subset' both use the same rho=0.25 IoU threshold, so Table 4's selection accuracy is evaluated under the same positive definition used to train the reward. That is a transparent conditioning choice, not a by-construction equality: 72.1% is a measured held-out value, and the paper explicitly states that proposal coverage and selection accuracy use different query sets and are not numerically composable (Sec. 4.4). Similarly, Table 3's use of an oracle best-view union is declared as a coverage diagnostic rather than an end-to-end ranking. The unreleased baseline reproductions (Fun3DU, TASA, AffordBot, Mask3D-F described as 'reproduced by us') are a reproducibility risk, not a circularity in the derivation. Overall, the pipeline's predictions are not equivalent to its inputs by construction, and the internal diagnostics are honestly labeled conditional measures.

Assumptions & free parameters 7 free parameters · 5 assumptions · 0 invented entities

The paper does not introduce new physical entities or exotic assumptions; it relies on standard ML machinery. The most notable ledger items are the rho=0.25 IoU threshold, which simultaneously defines training targets and evaluation subsets, and the assumption that VLM-parsed affordance types are correct at inference. Several hyperparameters are tuned but their values are not fully disclosed.

free parameters (7)
  • APG heatmap threshold gamma = not specified
    Binarizes the affordance heatmap before DBSCAN proposal extraction; controls the precision-recall tradeoff of APG proposals.
  • DBSCAN eps / min_samples = not specified
    Determines disconnected-component extraction and variable proposal count for small affordance regions.
  • IoU threshold rho = 0.25
    Defines positive proposals, the APG-covered subset, and the VPAR reward target set. Both module-level headline numbers (77.5 R@25 and 72.1 selection accuracy) are conditional on this value.
  • fusion threshold eta = 0.3
    Converts weighted multi-view voting scores into the final 3D mask; affects AP25/AP50.
  • reward format weight lambda_fmt = not stated
    Balances grounding reward against output-format reward in Eq. (2); ablated but exact value not reported.
  • GRPO KL coefficient beta = 5e-3
    KL-regularization strength in Eq. (5), a standard RL hyperparameter not justified by data.
  • number of retained views K = 25
    Fixed temporal neighborhood used after VLM view pruning; affects computation and multi-view fusion.
assumptions (5)
  • domain assumption Depth-tolerant z-buffer with local-plane disambiguation correctly determines which 3D annotated points are visible in each RGB view.
    APG's 2D supervision Y_a^v is generated by projecting 3D masks through this visibility index; if the projection is wrong, both training and lifted evaluation targets inherit the error. See Section 3.2, 'Data Preparation from 3D Labels'.
  • domain assumption The VLM-parsed affordance type a* is correct at inference time.
    APG is trained with ground-truth affordance type but at inference receives the parsing VLM's prediction; a wrong type removes the interaction region from the heatmap search space. See Section 3.2, 'Architecture and Training'.
  • domain assumption Point-level IoU between lifted 2D proposals and 3D ground-truth masks is a valid and sufficient reward signal for proposal selection.
    Eqs. (2)-(4) reward VPAR by 3D overlap; this assumes the metric used for training aligns with task success and that rho=0.25 is an adequate positive threshold.
  • ad hoc to paper Queries where no APG proposal reaches rho can be excluded from VPAR training without biasing the policy.
    Section 3.3 excludes APG coverage failures from VPAR training because 'no valid selection target exists'; the trained selector is never optimized to recover from proposal misses, yet end-to-end evaluation includes those queries.
  • domain assumption SceneFun3D's training split provides consistent alignment between RGB video views, camera parameters, and reconstructed point clouds.
    All lifting, visibility masking, and reward computation rely on this alignment; cross-dataset drops on MultiScan AP50 suggest reconstruction-domain mismatch remains. See Sections 3.2 and 4.5.

how reviews work

0 comments
Cite this review

Pith. "Pith review of ThinkAfford: Affordance-Centric Reasoning for Fine-Grained 3D Grounding in Cluttered Scenes." pith.science (2026). https://pith.science/paper/4WPO7KBJ

@misc{pith2026260810981,
  author       = {Pith},
  title        = {Pith review of: ThinkAfford: Affordance-Centric Reasoning for Fine-Grained 3D Grounding in Cluttered Scenes},
  year         = {2026},
  howpublished = {\url{https://pith.science/paper/4WPO7KBJ}},
  note         = {Machine review of arXiv:2608.10981}
}
read the original abstract

Task-driven 3D affordance grounding aims to localize the functional region in a cluttered 3D scene that enables an action specified by a natural-language instruction. Existing methods either predict 3D masks directly or construct them by selecting and fusing intermediate 2D/3D regions. However, they remain vulnerable to two intertwined failure modes: the predicted or selected regions may miss the target interaction area or have unsuitable granularity, while language grounding may confuse visually similar alternatives under relational instructions. To this end, we introduce ThinkAfford, which decouples high-recall affordance proposal generation from instruction-grounded reasoning. Specifically, the Affordance Proposal Generation module first uses learnable affordance prompts and multi-level visual features to predict interaction-conditioned heatmaps, extracting a variable number of fine-grained proposals without parsed object or part names as segmentation prompts. Visual-Prompted Affordance Reasoning then reasons over labeled proposal overlays using the full instruction, returning identifiers in a structured "think-then-answer" response. Moreover, Group Relative Policy Optimization uses proposal-level rewards from lifted 3D overlap to align VPAR selection with final 3D grounding. On the SceneFun3D validation split, ThinkAfford achieves 10.69% AP50 and 25.46% AP25 under the official evaluator, outperforming comparable 3D open-vocabulary and vision-language-model-based 2D-to-3D baselines. Module-level diagnostics further show that APG attains 77.5% recall at 25% intersection-over-union, while GRPO-trained VPAR achieves 72.1% selection accuracy on APG-covered queries, compared with 63.4% under supervised fine-tuning.

Figures

Figures reproduced from arXiv: 2608.10981 by the authors.

Figure 1
Figure 1. Overview of ThinkAfford. The method keeps a compact set of views, generates affordance-centric proposals, selects [PITH_FULL_IMAGE:figures/full_fig_p003_1.png] view at source ↗
Figure 2
Figure 2. An illustration of the APG module. Frozen [PITH_FULL_IMAGE:figures/full_fig_p003_2.png] view at source ↗
Figure 3
Figure 3. GRPO training for VPAR. Given an instruction and a labeled proposal overlay, the trainable VLM samples a group of [PITH_FULL_IMAGE:figures/full_fig_p004_3.png] view at source ↗
Figures from the paper (3 more)
Figure 4
Figure 4. Figure 4: Qualitative comparison. Columns compare four baselines and ThinkAfford with ground truth on repeated-part disambiguation and functional-region localization. Fun3dU Ours Scannet-Afford: “Plug the device in one of the sockets near the foot of the bed.” 3RScan-Afford: “Se…
Figure 5
Figure 5. Figure 5: Qualitative zero-shot cross-dataset transfer. Predictions on the four unseen datasets in [PITH_FULL_IMAGE:figures/full_fig_p006_5.png]
Figure 6
Figure 6. Figure 6: Qualitative real-world deployment. Five manipulation examples with scene-level robot approach views (top) and close-up execution views of the localized interaction regions (bottom) [PITH_FULL_IMAGE:figures/full_fig_p007_6.png]

Discussion (0). Continue with ORCID to comment.

Reference graph

Works this paper leans on

63 extracted references · 30 canonical work pages

  1. [1]

    Scaling Learning Algorithms Towards

    Bengio, Yoshua and LeCun, Yann , booktitle =. Scaling Learning Algorithms Towards

  2. [2]

    and Osindero, Simon and Teh, Yee Whye , journal =

    Hinton, Geoffrey E. and Osindero, Simon and Teh, Yee Whye , journal =. A Fast Learning Algorithm for Deep Belief Nets , volume =

  3. [3]

    2016 , publisher=

    Deep learning , author=. 2016 , publisher=

    2016

  4. [4]

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

    Scenefun3d: Fine-grained functionality and affordance understanding in 3d scenes , author=. Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition , pages=

    work page

  5. [5]

    Proceedings of the Computer Vision and Pattern Recognition Conference , pages=

    Functionality understanding and segmentation in 3D scenes , author=. Proceedings of the Computer Vision and Pattern Recognition Conference , pages=

    work page

  6. [6]

    arXiv preprint arXiv:2508.06206 , year=

    Affordance-R1: Reinforcement Learning for Generalizable Affordance Reasoning in Multimodal Large Language Model , author=. arXiv preprint arXiv:2508.06206 , year=

    arXiv

  7. [8]

    Segment Any 3D-Part in a Scene from a Sentence

    Segment Any 3D-Part in a Scene from a Sentence , author=. arXiv preprint arXiv:2506.19331 , year=

    work page Pith review arXiv

  8. [9]

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

    One-shot open affordance learning with foundation models , author=. Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition , pages=

    work page

Show all 63 references
  1. [10]

    International Journal of Computer Vision , volume=

    Learning to prompt for vision-language models , author=. International Journal of Computer Vision , volume=. 2022 , publisher=

    2022

  2. [11]

    arXiv preprint arXiv:2402.03300 , year=

    Deepseekmath: Pushing the limits of mathematical reasoning in open language models , author=. arXiv preprint arXiv:2402.03300 , year=

    arXiv

  3. [12]

    arXiv preprint arXiv:2304.07193 , year=

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

    arXiv

  4. [13]

    International conference on machine learning , pages=

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

    2021

  5. [14]

    arXiv preprint arXiv:2404.05164 , year=

    Rendering-enhanced automatic image-to-point cloud registration for roadside scenes , author=. arXiv preprint arXiv:2404.05164 , year=

    arXiv

  6. [15]

    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=

  7. [16]

    2023 IEEE International Conference on Robotics and Automation (ICRA) , pages=

    Mask3d: Mask transformer for 3d semantic instance segmentation , author=. 2023 IEEE International Conference on Robotics and Automation (ICRA) , pages=. 2023 , organization=

    2023

  8. [17]

    arXiv preprint arXiv:2306.13631 , year=

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

    arXiv

  9. [18]

    Proceedings of the Computer Vision and Pattern Recognition Conference , pages=

    Open-vocabulary functional 3d scene graphs for real-world indoor spaces , author=. Proceedings of the Computer Vision and Pattern Recognition Conference , pages=

  10. [19]

    arXiv e-prints , pages=

    Molmo and pixmo: Open weights and open data for state-of-the-art multimodal models , author=. arXiv e-prints , pages=

  11. [20]

    arXiv preprint arXiv:2506.17545 , year=

    Scene-R1: Video-Grounded Large Language Models for 3D Scene Reasoning without 3D Annotations , author=. arXiv preprint arXiv:2506.17545 , year=

    arXiv

  12. [21]

    Proceedings of the AAAI Conference on Artificial Intelligence , volume=

    Task-aware 3d affordance segmentation via 2d guidance and geometric refinement , author=. Proceedings of the AAAI Conference on Artificial Intelligence , volume=

  13. [22]

    Advances in Neural Information Processing Systems , volume=

    Affordbot: 3d fine-grained embodied reasoning via multimodal large language models , author=. Advances in Neural Information Processing Systems , volume=

  14. [23]

    arXiv preprint arXiv:2503.07909 , year=

    FunGraph: Functionality Aware 3D Scene Graphs for Language-Prompted Scene Interaction , author=. arXiv preprint arXiv:2503.07909 , year=

    arXiv

  15. [24]

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

    Laso: Language-guided affordance segmentation on 3d object , author=. Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition , pages=

  16. [25]

    CoRL , year=

    VLM-Grounder: A VLM Agent for Zero-Shot 3D Visual Grounding , author=. CoRL , year=

  17. [26]

    arXiv preprint arXiv:2502.20041 , year=

    3d-affordancellm: Harnessing large language models for open-vocabulary affordance detection in 3d worlds , author=. arXiv preprint arXiv:2502.20041 , year=

    arXiv

  18. [27]

    arXiv preprint arXiv:2505.04965 , year=

    Densegrounding: Improving dense language-vision semantics for ego-centric 3d visual grounding , author=. arXiv preprint arXiv:2505.04965 , year=

    arXiv

  19. [28]

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

    Graco: Granularity-controllable interactive segmentation , author=. Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition , pages=

  20. [29]

    arXiv preprint arXiv:1906.05797 , year=

    The replica dataset: A digital replica of indoor spaces , author=. arXiv preprint arXiv:1906.05797 , year=

    1906 arXiv

  21. [30]

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

    Scannet++: A high-fidelity dataset of 3d indoor scenes , author=. Proceedings of the IEEE/CVF International Conference on Computer Vision , pages=

  22. [31]

    arXiv preprint arXiv:2502.13923 , year=

    Qwen2.5-VL Technical Report , author=. arXiv preprint arXiv:2502.13923 , year=

    arXiv

  23. [32]

    arXiv preprint arXiv:2409.12191 , year=

    Qwen2-VL: Enhancing Vision-Language Model's Perception of the World at Any Resolution , author=. arXiv preprint arXiv:2409.12191 , year=

    arXiv

  24. [33]

    arXiv preprint arXiv:2308.12966 , year=

    Qwen-VL: A Versatile Vision-Language Model for Understanding, Localization, Text Reading, and Beyond , author=. arXiv preprint arXiv:2308.12966 , year=

    arXiv

  25. [34]

    Advances in neural information processing systems , volume=

    Multiscan: Scalable rgbd scanning for 3d environments with articulated objects , author=. Advances in neural information processing systems , volume=

  26. [35]

    arXiv preprint arXiv:2111.08897 , year=

    Arkitscenes: A diverse real-world dataset for 3d indoor scene understanding using mobile rgb-d data , author=. arXiv preprint arXiv:2111.08897 , year=

    arXiv

  27. [36]

    arXiv preprint arXiv:1709.06158 , year=

    Matterport3d: Learning from rgb-d data in indoor environments , author=. arXiv preprint arXiv:1709.06158 , year=

    arXiv

  28. [37]

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

    Scannet: Richly-annotated 3d reconstructions of indoor scenes , author=. Proceedings of the IEEE conference on computer vision and pattern recognition , pages=

  29. [38]

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

    Scanqa: 3d question answering for spatial scene understanding , author=. proceedings of the IEEE/CVF conference on computer vision and pattern recognition , pages=

  30. [39]

    Advances in neural information processing systems , volume=

    Visual instruction tuning , author=. Advances in neural information processing systems , volume=

  31. [40]

    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

  32. [41]

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

    Manipllm: Embodied multimodal large language model for object-centric robotic manipulation , author=. Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition , pages=

  33. [42]

    Proceedings of the Computer Vision and Pattern Recognition Conference , pages=

    Video-3d llm: Learning position-aware video representation for 3d scene understanding , author=. Proceedings of the Computer Vision and Pattern Recognition Conference , pages=

  34. [43]

    arXiv preprint arXiv:2210.03105 , year=

    Mask3d: Mask transformer for 3d semantic instance segmentation , author=. arXiv preprint arXiv:2210.03105 , year=

    arXiv

  35. [44]

    2023 IEEE/RSJ International Conference on Intelligent Robots and Systems (IROS) , pages=

    Open-vocabulary affordance detection in 3d point clouds , author=. 2023 IEEE/RSJ International Conference on Intelligent Robots and Systems (IROS) , pages=. 2023 , organization=

    2023

  36. [45]

    Conference on Robot Learning , pages=

    One-shot transfer of affordance regions? affcorrs! , author=. Conference on Robot Learning , pages=. 2023 , organization=

    2023

  37. [46]

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

    Demo2vec: Reasoning object affordances from online videos , author=. Proceedings of the IEEE Conference on Computer Vision and Pattern Recognition , pages=

  38. [47]

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

    Cross-view aggregation network for stereo image super-resolution , author=. Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition , pages=

  39. [48]

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

    Learning affordance grounding from exocentric images , author=. Proceedings of the IEEE/CVF conference on computer vision and pattern recognition , pages=

  40. [49]

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

    Locate: Localize and transfer object parts for weakly supervised affordance grounding , author=. Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition , pages=

  41. [50]

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

    Understanding 3d object interaction from a single image , author=. Proceedings of the IEEE/CVF International Conference on Computer Vision , pages=

  42. [51]

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

    Affordancellm: Grounding affordance from vision language models , author=. Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition , pages=

  43. [52]

    Neural Computing and Applications , volume=

    Object affordance detection with relationship-aware network , author=. Neural Computing and Applications , volume=. 2020 , publisher=

    2020

  44. [53]

    2018 IEEE international conference on robotics and automation (ICRA) , pages=

    Affordancenet: An end-to-end deep learning approach for object affordance detection , author=. 2018 IEEE international conference on robotics and automation (ICRA) , pages=. 2018 , organization=

    2018

  45. [54]

    IEEE Transactions on Neural Networks and Learning Systems , year=

    Learning visual affordance grounding from demonstration videos , author=. IEEE Transactions on Neural Networks and Learning Systems , year=

  46. [55]

    Proceedings of the AAAI Conference on Artificial Intelligence , volume=

    Learning 2d invariant affordance knowledge for 3d affordance grounding , author=. Proceedings of the AAAI Conference on Artificial Intelligence , volume=

  47. [56]

    International Journal of Computer Vision , volume=

    Grounded affordance from exocentric view , author=. International Journal of Computer Vision , volume=. 2024 , publisher=

    2024

  48. [57]

    2024 IEEE International Conference on Robotics and Automation (ICRA) , pages=

    Language-conditioned affordance-pose detection in 3d point clouds , author=. 2024 IEEE International Conference on Robotics and Automation (ICRA) , pages=. 2024 , organization=

    2024

  49. [58]

    2024 IEEE International Conference on Robotics and Automation (ICRA) , pages=

    Open-vocabulary affordance detection using knowledge distillation and text-point correlation , author=. 2024 IEEE International Conference on Robotics and Automation (ICRA) , pages=. 2024 , organization=

    2024

  50. [59]

    European Conference on Computer Vision , pages=

    Agent3d-zero: An agent for zero-shot 3d understanding , author=. European Conference on Computer Vision , pages=. 2024 , organization=

    2024

  51. [60]

    EvalAI: Towards Better Evaluation Systems for AI Agents , author =

  52. [61]

    arXiv:2304.02643 , year=

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

    arXiv

  53. [62]

    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=

    arXiv

  54. [63]

    2024 , eprint=

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

    2024

  55. [64]

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

    Rio: 3d object instance re-localization in changing indoor environments , author=. Proceedings of the IEEE/CVF International Conference on Computer Vision , pages=

Pith tools

Reviewed August 12, 2026 · model on record in the stance chip above.