Pith. sign in

REVIEW 4 major objections 7 minor 1 cited by

MonoGlass3D: Monocular 3D Glass Detection with Plane Regression and Adaptive Feature Fusion

T0 review · 4 major / 7 minor · reviewed 2026-08-15 · deepseek-v4-flash

Pith's one-line read Glass is nearly invisible to depth sensors because light passes through it; MonoGlass3D claims the fix is to stop estimating depth pixel-by-pixel and instead regress the 3D plane each glass surface lies on, backed by a new 1,437-frame…

desk verdict Solid dataset and method, but the depth labels are plane fits to manually placed boxes, not measured glass depths, so the depth claims are weaker than the segmentation ones. read the letter →

arxiv 2509.05599 v1 pith:CWS6XWXY submitted 2025-09-06 cs.RO

classification cs.RO
keywords glassdetectionmonocular3Dplaneparameterregressiondepthestimationadaptivefeaturefusioncenternessdistancelosstransparentsurfaces
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 establish that monocular 3D glass detection should be formulated as plane parameter regression with centerness-guided adaptive fusion, rather than as direct pixel-by-pixel depth estimation. It contributes a new real-world dataset of 1,437 annotated frames from 50 scenes, with up to ten glass planes per image, depth ranges from 0.07 m to 17.76 m, and 21 night scenes, where depth labels come from projecting 2D masks onto planes fitted to LiDAR-derived 3D boxes. On this dataset, MonoGlass3D reports the best depth metrics among the compared methods, including general-purpose monocular depth models, using 52.4M parameters, and it reports state-of-the-art glass segmentation on the GDD and GSD benchmarks when the plane branch is switched off. A sympathetic reader would take the core claim to be that the planarity of glass is a stronger cue than any appearance-based context, and that exploiting it explicitly is what makes transparent surfaces tractable for a single RGB camera.

What carries the argument

The load-bearing mechanism is the plane parameterization of glass. Each glass surface is represented by $(\theta_1, \theta_2, d)$, where $\theta_1,\theta_2$ are polar angles of the unit normal restricted to the $-z$ hemisphere and $d$ is the plane intercept; this reduces the natural four-variable plane representation to three parameters and removes the need for post-hoc normalization. Predicted plane parameters are converted into depth by projecting each pixel onto its plane, $depth = d / (n^T K^{-1}[u_x, u_y, 1]^T)$, so depth inherits the geometric regularity of the plane. Training is driven by a plane distance loss, which places four equally spaced points around each predicted pixel's 3D position and sums their distances to the ground-truth plane; unlike a depth loss, this error is invariant under viewpoint and plane orientation. The second mechanism is centerness-based adaptive fusion: centerness $C=\sqrt{d_{\min}/d_{\max}}$ is predicted early, and features entering the segmentation and plane heads are multiplied by $(1-C)$, steering the network toward the context-rich region around glass boundaries; the cascade repeats this at three backbone scales, with self- and cross-attention refining the final plane and segmentation features.

What would settle it

Take a handful of new scenes with glass, measure the true glass-surface positions independently of LiDAR-through-glass — for example with a structured-light scan, a rangefinder placed flat against the pane, or direct tape measurement — and compare MonoGlass3D's predicted planes and its training labels against those measurements. If the plane-fitted labels and the network's predictions agree with each other but disagree with the independent measurements by the same margin as the reported errors, then the plane-fit labeling pipeline, not the network, is responsible for the good numbers.

Watch

Extended reading notes

Core claim

On the paper's own terms, the central discovery is that the transparency of glass stops being the main obstacle once the network is constrained to see glass as 3D planes. The paper shows that regressing three plane parameters — two polar angles for the normal and one intercept — and projecting them to depth yields flatter, more accurate depth on glass surfaces than direct depth regression, and that reweighting features by $1-C$, where $C=\sqrt{d_{\min}/d_{\max}}$ is a per-pixel centerness, provides the contextual signal that appearance cues like boundaries and reflections cannot. The supporting evidence is a new dataset whose annotation pipeline produces dense depth labels from sparse 3D plane fits, and evaluations in which MonoGlass3D achieves the best reported depth error on all metrics, with IoU of 0.920 on GDD and 0.872 on GSD for segmentation and stronger held-out-scene depth than the baselines.

Load-bearing premise

The depth labels that train and evaluate everything are produced by fitting planes to manual 3D boxes on a LiDAR-visual SLAM point cloud, which assumes the point cloud actually contains points on the glass surface rather than behind it — an assumption the paper itself concedes is not directly observable.

Editorial extensions

If this is right

  • A robot with only an RGB camera could obtain glass depth accurate enough for navigation and manipulation, sidestepping the LiDAR and depth cameras that see through or ignore transparent surfaces.
  • Segmentation and 3D localization collapse into one forward pass: the same features produce masks and plane parameters, so 'where the glass is' and 'where it sits in 3D' arrive together.
  • Coplanar glass instances — several windows on one wall — will be reconstructed as mutually consistent flat surfaces because they share plane parameters, removing the noisy interior depths that direct regression produces on featureless glass.
  • Since the plane distance loss is orientation-invariant, training is not dominated by frontoparallel planes, and oblique or multi-angle glass configurations contribute equally to the gradient.
  • The new dataset, with up to ten planes per image and night-time illumination, raises the ceiling for evaluating 3D glass perception beyond the one-or-two-plane datasets used previously.

Reading between the lines

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

  • Our inference: the same plane-regression recipe should transfer to other planar transparent or reflective surfaces — mirrors, display cases, acrylic partitions — because they share the same failure mode of letting light and depth-sensor signals pass through; the paper does not test these.
  • Our inference: the plane distance loss could work as a geometric regularizer for general monocular depth estimation in piecewise-planar scenes, not just glass; the paper's ablation shows it beats depth supervision in this architecture, but the broader transfer is unverified.
  • Our inference: fusing the predicted plane intercepts with sparse LiDAR returns that do hit glass (available in some frames) could resolve monocular scale ambiguity and is a direct, testable extension the paper leaves for future work.
  • Our inference: the $(1-C)$ centerness weighting may benefit detection of any object whose interior is less informative than its boundary — windows, mirrors, fences — which would make the mechanism useful beyond glass.
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

4 major / 7 minor

Summary. The paper presents MonoGlass3D, a monocular 3D glass detection method, together with a new real-world dataset. The dataset contains 1,437 annotated frames from 50 scenes, with glass segmentation masks, 3D plane parameters, and depth maps. Depth labels are generated by manually placing thin 3D bounding boxes on a FAST-LIVO2 point-cloud map, fitting a plane to each box, and projecting 2D masks onto those planes. The proposed network uses a DINOv2 backbone with a centerness-based adaptive feature fusion module, a plane regression head that predicts angles and intercept, and attention layers, trained with segmentation, centerness, and plane losses. Experiments report state-of-the-art glass segmentation on GDD and GSD, and the best depth estimation accuracy on the new dataset and on a held-out scene split, while using fewer parameters than the largest baselines. An ablation study isolates the contributions of adaptive fusion, attention, cascades, and the proposed plane distance loss.

Significance. If the depth ground truth is trustworthy, the dataset is a useful contribution for robotic perception of transparent surfaces, and the plane regression formulation with a viewpoint-invariant plane distance loss is a well-motivated alternative to direct depth regression. The segmentation results are validated on external public benchmarks (GDD and GSD), which lends credibility to the architecture's representational strength. The paper also provides clear ablations and promises code and dataset release. However, the significance of the core depth claims is conditional on the validity of the synthesized plane-based depth labels, and the current manuscript does not provide independent evidence for that validity.

major comments (4)
  1. [Section III-C, Eqs. (1)-(4); Section V-F] The depth ground truth is not a direct measurement: it is constructed by fitting planes to 3D boxes manually placed on a LiDAR point-cloud map, and Section V-F concedes that 'the 3D positions of glass are not directly observable'. With the LiDAR operated in single-first-return mode, returns may originate from objects behind the glass, from the frame, or from refracted paths, so the fitted plane is an annotation assumption rather than a range measurement. Because MonoGlass3D outputs plane parameters and is trained and evaluated on depths projected from exactly these fitted planes, the depth margins over baselines in Tables I and V (e.g., 0.063 vs. 0.067 Abs Rel on the full set) could reflect structural alignment with the label-generation model rather than physical localization accuracy. I request an independent validation of a sample of depth labels, for example by comparing predicted plane distances against hand-measured distances or known architectural plans, or by providing point-cloud evidence that first returns lie on the glass surface itself.
  2. [Section V-C.2, Tables II and III] The GDD and GSD segmentation comparisons are performed with plane regression layers disabled, yet the text presents these results as 'our model achieves state-of-the-art performance' and attributes them to MonoGlass3D. Table IV shows that the full architecture with cascades and plane losses changes IoU on the in-house split (e.g., IoU 0.925 without cascades vs. 0.942 with them). The reader cannot tell whether the segmentation gains come from the full MonoGlass3D design or from a segmentation-only variant. Please state explicitly which components are disabled for these experiments and justify that the variant is representative of the proposed method; otherwise the external segmentation claim tests a different model.
  3. [Section IV-B, Eq. (5)] The angular parameterization appears not to cover all plane normals in the -z hemisphere. For a normal with z<0 and x<0, Eq. (5) gives θ2 = -arccos(z/r_xz); as z/r_xz approaches -1, this value approaches -π, which is outside the claimed range [-π/2, π/2] that the text uses for tanh up-scaling. The case r_xz=0 is also undefined, which would affect horizontal glass surfaces such as ceilings or floors. This means certain plane orientations cannot be represented by the network's output range. Please clarify the intended angular range, correct Eq. (5) if needed, and discuss singularity handling.
  4. [Section V-E, Table V] The held-out scene experiment uses six scenes from the same dataset, and their labels inherit the same plane-fit assumption as the training data. As a generalization test, this setup cannot detect systematic bias in the label-generation process; it only checks generalization to new scenes under the same annotation protocol. Please add results on an independently measured scene (e.g., with a different depth sensor or manual distance measurements) or, at minimum, provide a quantitative analysis of label uncertainty across the annotation pipeline.
minor comments (7)
  1. [Abstract and Section V-F] The abstract claims 'precise 3D annotations', but Section V-F states that 3D positions of glass are not directly observable; please temper the abstract or clarify that precision is relative to the plane-fit assumption.
  2. [Section II-B] There is a typo in the sentence 'GWDepth [11] uses a depth camera to capture the depth values at glass boundaries and and interpolate pixel-wise glass depths' – 'and and' should be 'and'.
  3. [Section III-C, Eq. (3)] The term 'interception' should be 'intercept' throughout the description of plane parameters d.
  4. [Section IV-B, Eq. (5)] The notation uses r both as the scalar magnitude (r=1 for unit vectors) and as part of r_xz; please use distinct symbols to avoid confusion.
  5. [Section IV-D, Eq. (8)] The loss term is denoted 'lp' in Eq. (8) but 'L_p' elsewhere; unify the notation.
  6. [Table IV] The ablation table is difficult to read because the checkmark columns are not labelled clearly for each row; please reformat it so that each configuration is unambiguous.
  7. [Section V-C.1] The phrase '52.4 Mtrainable parameters' is missing a space; it should read '52.4 M trainable parameters'.

Circularity Check

0 steps flagged · score 2.0 of 10

No load-bearing circularity: the depth benchmark is plane-annotated by design (Sec. III-C), but the network is trained and tested on held-out data and segmentation is validated on external datasets; the only self-citation is contextual.

full rationale

The paper's central derivation chain is not circular. The depth ground truth is produced by an independent annotation pipeline: annotators place 3D boxes in a LiDAR map, fit planes via least squares (Eq. 3), and project each mask onto the plane (Eq. 4). These labels are generated before training and are not produced by the network; the method is then evaluated on held-out scenes (Table V) excluded from training, so no fitted parameter is being renamed as a prediction. The plane-regression output is converted to depth with the same projection formula used to create the labels, which is a consistent benchmark design rather than a reduction of the prediction to its input: a direct depth-regression baseline is trained on the same plane-projected labels and still underperforms (Table I). Segmentation claims are independently supported by external GDD and GSD benchmarks (Tables II-III). The paper explicitly acknowledges the dataset's main limitation in Sec. V-F: 'the 3D positions of glass are not directly observable', which is a validity caveat about the LiDAR-based labels, not circular reasoning. The only apparent self-citation is [3], a co-authored LiDAR-camera fusion paper used for the contextual claim about safe autonomous operation; it is not load-bearing. No uniqueness theorem or ansatz is imported from the authors' prior work, and the plane prior is justified geometrically in Secs. II-C and IV-B. Overall circularity burden is minimal (2/10).

Assumptions & free parameters 2 free parameters · 4 assumptions · 0 invented entities

No new physical entities are introduced; the contributions are a dataset, a network, and a loss. The main unverified inputs are domain assumptions about the dataset's ground truth and the LiDAR-derived annotation process.

free parameters (2)
  • Stage loss weights = 0.1, 0.1, 0.2, 0.6 for Lp and Ls; 0.2, 0.3, 0.5 for Lc
    Hand-set weights in Eq. (12) that balance losses across the four cascade stages; these affect the final performance numbers reported in the ablation and comparisons.
  • Intercept upscale factor = 5
    The predicted plane intercept d is multiplied by 5 (Sec. IV-B) to scale the regression target; chosen by hand and used in all experiments.
assumptions (4)
  • domain assumption All glass in the dataset can be represented as 3D planes fit to bounding boxes
    The entire annotation pipeline (Sec. III-C) and the depth ground truth (Eq. 4) rest on this; Sec. V-F admits curved glass is not handled.
  • domain assumption LiDAR first-return point clouds contain reliable 3D points on glass surfaces for manual annotation
    Sec. III-C places glass bounding boxes on the point cloud map; Sec. V-F notes 3D glass positions are not directly observable, yet no validation against other sensors is provided.
  • domain assumption FAST-LIVO2 provides accurate camera poses and point cloud maps
    The annotations and plane fits assume the SLAM output used in Sec. III-B is accurate enough for plane labeling; drift or scale error would bias ground truth.
  • standard math The polar conversion in Eq. (5) fully parameterizes plane normals with no practical singularities
    Spherical-like parameterization of the lower hemisphere; singularities near r_xz=0 are not discussed, but appear manageable in the reported range.

how reviews work

0 comments
Cite this review

Pith. "Pith review of MonoGlass3D: Monocular 3D Glass Detection with Plane Regression and Adaptive Feature Fusion." pith.science (2026). https://pith.science/paper/CWS6XWXY

@misc{pith2026250905599,
  author       = {Pith},
  title        = {Pith review of: MonoGlass3D: Monocular 3D Glass Detection with Plane Regression and Adaptive Feature Fusion},
  year         = {2026},
  howpublished = {\url{https://pith.science/paper/CWS6XWXY}},
  note         = {Machine review of arXiv:2509.05599}
}
read the original abstract

Detecting and localizing glass in 3D environments poses significant challenges for visual perception systems, as the optical properties of glass often hinder conventional sensors from accurately distinguishing glass surfaces. The lack of real-world datasets focused on glass objects further impedes progress in this field. To address this issue, we introduce a new dataset featuring a wide range of glass configurations with precise 3D annotations, collected from distinct real-world scenarios. On the basis of this dataset, we propose MonoGlass3D, a novel approach tailored for monocular 3D glass detection across diverse environments. To overcome the challenges posed by the ambiguous appearance and context diversity of glass, we propose an adaptive feature fusion module that empowers the network to effectively capture contextual information in varying conditions. Additionally, to exploit the distinct planar geometry of glass surfaces, we present a plane regression pipeline, which enables seamless integration of geometric properties within our framework. Extensive experiments demonstrate that our method outperforms state-of-the-art approaches in both glass segmentation and monocular glass depth estimation. Our results highlight the advantages of combining geometric and contextual cues for transparent surface understanding.

Figures

Figures reproduced from arXiv: 2509.05599 by the authors.

Figure 1
Figure 1. Challenges and our proposed method. (a) Existing 3D glass dataset [11] only contains simple glass structures with one or two planes. Synthetic datasets suffers from domain gap problems, which makes the models hard to generalize in real-world scenarios, as depicted by depth map from Depth Anything V2 Metric [12] (pretrained on Hypersim [13]). (b) Our dataset is collected from complex glass structures in real-world. (… view at source ↗
Figure 2
Figure 2. Proposed dataset construction procedures. (a) Data collection device, we integrated a LiDAR and a camera with a wheel-legged robot. (b) Collected scenes with complex glass structures. (c) 3D glass data annotation procedures. PlaneNet [24] introduces the first end-to-end neural network for piecewise planar reconstruction from a single RGB image, they estimate plane parameters and segmentation masks to yield structure… view at source ↗
Figure 3
Figure 3. Statistics of proposed dataset. (a) Distribution of number of planes in individual images. (b) Distribution of the range of glass depths in individual images. The proposed labeling pipeline enables efficient generation of precise depth ground truth for a large number of images with minimal manual effort. Building on this capability, our dataset exhibits a substantially greater diversity in glass planar configuration… view at source ↗
Figures from the paper (8 more)
Figure 4
Figure 4. Figure 4: Proposed MonoGlass3D framework. Our network consists DINOv2 backbone, centerness adaptive fusion module, and attention layers. Our network jointly performs glass segmentation and plane regression, which are supervised by segmentation loss Ls, centerness loss Lc, and pl…
Figure 5
Figure 5. Figure 5: Coordinate conversion of plane normal vector. Cartesian vector (x, y, z) is mapped to its corresponding polar coordinate (θ1, θ2). vector and one for the intercept), our angular representation reduces plane estimation to three parameters, [θ1 θ2 d], and eliminates the …
Figure 6
Figure 6. Figure 6: Centerness map examples. Image with masks (left) and correspond￾ing centerness maps (right). context is therefore a crucial property for both plane estima￾tion and glass segmentation. Previous works have leveraged boundary information [4], [5] to enhance contextual lea…
Figure 7
Figure 7. Figure 7: Drawbacks of depth loss and proposed plane distance loss. (a) Depth error at same angular error varies largely depending on plane angles. (b) Plot of depth error on the plane angle domain. (c) Plane distance loss, measures the distance between 3D planes (white dotted l…
Figure 8
Figure 8. Figure 8: Comparisons of glass detection results. (a) Coplanar structures contain glass surfaces lying on same plane. (b) Multi-angle surfaces contain scenes with glass instances with relatively small angular difference. (c) Multi￾occluded layers contain scenes with overlapping …
Figure 9
Figure 9. Figure 9: Comparisons of glass segmentation results. Compared to previous approaches, our method is better at accurately detecting glass in images, even when the surrounding context is ambiguous. Anything V2 (ViT-B). To facilitate a more intuitive compari￾son, we also visualize …
Figure 10
Figure 10. Figure 10: Attention maps with and without AF. The attention maps at the red dots show that AF modules enable the network to recognize different planes. model comprising only the backbone and prediction heads. Subsequently, we incrementally add the cross-attention (CA) module, s…
Figure 11
Figure 11. Figure 11: 3D glass detection results in real scenes. We evaluate our approach in scenes that are absent in the training set, and it maintains strong performance. does not always hold in real-world scenarios. In practice, glass surface shapes follow a long-tailed distribution: w…

Discussion (0). Continue with ORCID to comment.

Forward citations

Cited by 1 Pith paper

Reviewed papers in the Pith corpus that reference this work. Sorted by Pith novelty score. Full citation record

  1. Enhancing Glass Surface Reconstruction via Depth Prior for Robot Navigation

    cs.RO 2026-04 unverdicted novelty 6.0 of 10

    Local RANSAC alignment of Depth Anything-style priors to raw sensor depth recovers metric glass-surface geometry without training, and the new GlassRecon benchmark shows large gains on severely corrupted hard samples.

Reference graph

Works this paper leans on

32 extracted references · 26 canonical work pages · cited by 1 Pith paper

  1. [1]

    Robotic perception of transparent objects: A review,

    J. Jiang, G. Cao, J. Deng, T.-T. Do, and S. Luo, “Robotic perception of transparent objects: A review,”IEEE Transactions on Artificial Intelligence, vol. 5, no. 6, pp. 2547–2567, 2023

  2. [2]

    Don’t hit me! glass detection in real-world scenes,

    H. Mei, X. Yang, Y . Wang, Y . Liu, S. He, Q. Zhang, X. Wei, and R. W. Lau, “Don’t hit me! glass detection in real-world scenes,” in Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition, 2020, pp. 3687–3696

  3. [3]

    Omnicolor: a global camera pose optimization approach of lidar-360camera fusion for colorizing point clouds,

    B. Liu, G. Zhao, J. Jiao, G. Cai, C. Li, H. Yin, Y . Wang, M. Liu, and P. Hui, “Omnicolor: a global camera pose optimization approach of lidar-360camera fusion for colorizing point clouds,” in2024 IEEE International Conference on Robotics and Automation, 2024, pp. 6396– 6402

  4. [4]

    Enhanced boundary learning for glass-like object segmen- tation,

    H. He, X. Li, G. Cheng, J. Shi, Y . Tong, G. Meng, V . Prinet, and L. Weng, “Enhanced boundary learning for glass-like object segmen- tation,” inProceedings of the IEEE/CVF International Conference on Computer Vision, 2021, pp. 15 859–15 868

  5. [5]

    Rfenet: Towards reciprocal feature evolution for glass segmentation,

    K. Fan, C. Wang, Y . Wang, C. Wang, R. Yi, and L. Ma, “Rfenet: Towards reciprocal feature evolution for glass segmentation,”arXiv preprint arXiv:2307.06099, 2023

  6. [6]

    Path planning in uncertain environment with moving obstacles using warm start cross entropy,

    X. Tao, N. Lang, H. Li, and D. Xu, “Path planning in uncertain environment with moving obstacles using warm start cross entropy,” IEEE/ASME Transactions on Mechatronics, vol. 27, no. 2, pp. 800–810, 2021

  7. [7]

    Where shall i touch? vision-guided tactile poking for transparent object grasping,

    J. Jiang, G. Cao, A. Butterworth, T.-T. Do, and S. Luo, “Where shall i touch? vision-guided tactile poking for transparent object grasping,” IEEE/ASME Transactions on Mechatronics, vol. 28, no. 1, pp. 233–244, 2022

  8. [8]

    Path planning for robotic manip- ulator in complex multi-obstacle environment based on improved rrt,

    L. Jiang, S. Liu, Y . Cui, and H. Jiang, “Path planning for robotic manip- ulator in complex multi-obstacle environment based on improved rrt,” IEEE/ASME Transactions on Mechatronics, vol. 27, no. 6, pp. 4774– 4785, 2022

Show all 32 references
  1. [9]

    A robot-assisted back-imaging measurement system for transparent glass,

    J. Xu, N. Xi, C. Zhang, Q. Shi, and J. Gregory, “A robot-assisted back-imaging measurement system for transparent glass,”IEEE/ASME Transactions on Mechatronics, vol. 17, no. 4, pp. 779–788, 2011

  2. [10]

    RGB-D local implicit function for depth completion of transparent objects,

    L. Zhu, A. Mousavian, Y . Xiang, H. Mazhar, J. van Eenbergen, S. Deb- nath, and D. Fox, “RGB-D local implicit function for depth completion of transparent objects,” inProceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition, 2021, pp. 4649–4658

  3. [11]

    Monocular depth estimation for glass walls with context: a new dataset and method,

    Y . Liang, B. Deng, W. Liu, J. Qin, and S. He, “Monocular depth estimation for glass walls with context: a new dataset and method,”IEEE Transactions on Pattern Analysis and Machine Intelligence, vol. 45, no. 12, pp. 15 081–15 097, 2023

  4. [12]

    Depth anything v2,

    L. Yang, B. Kang, Z. Huang, Z. Zhao, X. Xu, J. Feng, and H. Zhao, “Depth anything v2,”Advances in Neural Information Processing Sys- tems, vol. 37, pp. 21 875–21 911, 2024

  5. [13]

    Hypersim: A photorealistic synthetic dataset for holistic indoor scene understanding,

    M. Roberts, J. Ramapuram, A. Ranjan, A. Kumar, M. A. Bautista, N. Paczan, R. Webb, and J. M. Susskind, “Hypersim: A photorealistic synthetic dataset for holistic indoor scene understanding,” inProceedings of the IEEE/CVF International Conference on Computer Vision, 2021, pp. 1...

  6. [14]

    Glass makes blurs: Learning the visual blurriness for glass surface detection,

    F. Qi, X. Tan, Z. Zhang, M. Chen, Y . Xie, and L. Ma, “Glass makes blurs: Learning the visual blurriness for glass surface detection,”IEEE Transactions on Industrial Informatics, vol. 20, no. 4, pp. 6631–6641, 2024. IEEE/ASME TRANSACTIONS ON MECHATRONICS 11

  7. [15]

    Polarmask: Single shot instance segmentation with polar rep- resentation,

    E. Xie, P. Sun, X. Song, W. Wang, X. Liu, D. Liang, C. Shen, and P. Luo, “Polarmask: Single shot instance segmentation with polar rep- resentation,” inProceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition, 2020, pp. 12 193–12 202

  8. [16]

    SAM 2: Segment anything in images and videos,

    N. Ravi, V . Gabeur, Y .-T. Hu, R. Hu, C. Ryali, T. Ma, H. Khedr, R. R ¨adle, C. Rolland, L. Gustafsonet al., “SAM 2: Segment anything in images and videos,”arXiv preprint arXiv:2408.00714, 2024

  9. [17]

    Rich context aggregation with reflection prior for glass surface detection,

    J. Lin, Z. He, and R. W. Lau, “Rich context aggregation with reflection prior for glass surface detection,” inProceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition, 2021, pp. 13 415–13 424

  10. [18]

    Multi-view dynamic reflection prior for video glass surface detection,

    F. Liu, Y . Liu, J. Lin, K. Xu, and R. W. Lau, “Multi-view dynamic reflection prior for video glass surface detection,” inProceedings of the AAAI Conference on Artificial Intelligence, vol. 38, no. 4, 2024, pp. 3594–3602

  11. [19]

    Exploiting semantic relations for glass surface detection,

    J. Lin, Y .-H. Yeung, and R. Lau, “Exploiting semantic relations for glass surface detection,”Advances in Neural Information Processing Systems, vol. 35, pp. 22 490–22 504, 2022

  12. [20]

    Deep polarization cues for transparent object segmenta- tion,

    A. Kalra, V . Taamazyan, S. K. Rao, K. Venkataraman, R. Raskar, and A. Kadambi, “Deep polarization cues for transparent object segmenta- tion,” inProceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition, 2020, pp. 8602–8611

  13. [21]

    Glass segmentation using intensity and spectral polarization cues,

    H. Mei, B. Dong, W. Dong, J. Yang, S.-H. Baek, F. Heide, P. Peers, X. Wei, and X. Yang, “Glass segmentation using intensity and spectral polarization cues,” inProceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition, 2022, pp. 12 622–12 631

  14. [22]

    Glass segmentation with RGB-Thermal image pairs,

    D. Huo, J. Wang, Y . Qian, and Y .-H. Yang, “Glass segmentation with RGB-Thermal image pairs,”IEEE Transactions on Image Processing, vol. 32, pp. 1911–1926, 2023

  15. [23]

    Clear grasp: 3D shape estimation of transparent objects for manipulation,

    S. Sajjan, M. Moore, M. Pan, G. Nagaraja, J. Lee, A. Zeng, and S. Song, “Clear grasp: 3D shape estimation of transparent objects for manipulation,” in2020 IEEE International Conference on Robotics and Automation, 2020, pp. 3634–3642

  16. [24]

    PlaneNet: Piece-wise planar reconstruction from a single RGB image,

    C. Liu, J. Yang, D. Ceylan, E. Yumer, and Y . Furukawa, “PlaneNet: Piece-wise planar reconstruction from a single RGB image,” inPro- ceedings of the IEEE Conference on Computer Vision and Pattern Recognition, 2018, pp. 2579–2588

  17. [25]

    Recovering 3D planes from a single image via convolutional neural networks,

    F. Yang and Z. Zhou, “Recovering 3D planes from a single image via convolutional neural networks,” inProceedings of the European Conference on Computer Vision, 2018, pp. 85–100

  18. [26]

    PlaneRCNN: 3D plane detection and reconstruction from a single image,

    C. Liu, K. Kim, J. Gu, Y . Furukawa, and J. Kautz, “PlaneRCNN: 3D plane detection and reconstruction from a single image,” inProceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recogni- tion, 2019, pp. 4450–4459

  19. [27]

    P3Depth: Monocular depth estimation with a piecewise planarity prior,

    V . Patil, C. Sakaridis, A. Liniger, and L. Van Gool, “P3Depth: Monocular depth estimation with a piecewise planarity prior,” inProceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition, 2022, pp. 1610–1621

  20. [28]

    FAST-LIVO2: Fast, direct lidar-inertial-visual odometry,

    C. Zheng, W. Xu, Z. Zou, T. Hua, C. Yuan, D. He, B. Zhou, Z. Liu, J. Lin, F. Zhuet al., “FAST-LIVO2: Fast, direct lidar-inertial-visual odometry,”IEEE Transactions on Robotics, 2024

  21. [29]

    FCOS: Fully convolutional one- stage object detection,

    Z. Tian, C. Shen, H. Chen, and T. He, “FCOS: Fully convolutional one- stage object detection,” inProceedings of the IEEE/CVF International Conference on Computer Vision, 2019, pp. 9627–9636

  22. [30]

    PyTorch: An imperative style, high-performance deep learning library,

    A. Paszke, S. Gross, F. Massa, A. Lerer, J. Bradbury, G. Chanan, T. Killeen, Z. Lin, N. Gimelshein, L. Antigaet al., “PyTorch: An imperative style, high-performance deep learning library,”Advances in Neural Information Processing Systems, vol. 32, 2019

  23. [31]

    Zoedepth: Zero-shot transfer by combining relative and metric depth,

    S. F. Bhat, R. Birkl, D. Wofk, P. Wonka, and M. M ¨uller, “Zoedepth: Zero-shot transfer by combining relative and metric depth,”arXiv preprint arXiv:2302.12288, 2023. Kai Zhang(Student Member, IEEE) Recieved the B.Eng. degree in mechatronics engineering from the University of ...

  24. [2018]

    From 2018 to 2021, he held several positions at the National University of Singapore; University College London, London, U.K.; University of Cali- fornia, Berkeley, Berkeley, CA, USA; and Harvard University, Cambridge, MA, USA. He is currently an Assistant Professor with the R...

Pith tools

Reviewed August 15, 2026 · model on record in the stance chip above.