Pith. sign in

REVIEW 4 major objections 6 minor 1 cited by

Regularized interpolation in 4D neural fields enables optimization of 3D printed geometries

T0 review · 4 major / 6 minor · reviewed 2026-08-09 · deepseek-v4-flash

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

Pith's one-line read Regularized 4D neural fields can predict how a 3D-printed part changes shape when the flow rate changes, and that prediction can be used to optimize each printed layer.

desk verdict A useful application with a real dataset, but the interpolation claim is thinner than the abstract suggests and the regularization is a zero-derivative prior, not Lipschitz. read the letter →

arxiv 2502.01517 v1 pith:CI6H5NTN submitted 2025-02-03 cs.GR cs.AI

Christos Margadji , Andi Kuswoyo , Sebastian W. Pattinson This is my paper · ORCID
classification cs.GRcs.AI
keywords 4Dneuralfieldsgradient-driveninterpolationregularizationmaterialextrusion3DprintinggeometricfidelityflowrateoptimizationsigneddistancefunctionsCTscandigitaltwinsprocessparameter
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 argues that a 4D neural field—three spatial coordinates plus a process parameter such as filament flow rate—can represent how a 3D-printed part changes shape as the flow rate changes, and that a new regularization term makes that representation reliable at flow rates never used in training. The regularization, called gradient-driven interpolation regularization (GDIR), penalizes the partial derivative of the network output with respect to the flow rate, encouraging small changes in flow to produce small, physically plausible changes in the reconstructed volume. With this field, the authors optimize the flow rate layer by layer for a printed bunny, showing that regions like the ears benefit from higher flow while the body benefits from slight under-extrusion. If correct, the approach turns a learned shape model into a simulator for process-parameter optimization, cutting trial-and-error printing and reducing waste.

What carries the argument

The load-bearing object is the 4D neural field $F_\Theta(x,y,z;\varphi)$ trained on registered CT scans of parts printed at nine flow rates, with the gradient-driven interpolation regularization (GDIR) as the mechanism that makes interpolation safe. GDIR adds $\lambda\,\|\partial F_\Theta/\partial \varphi\|_2^2$ to the mean-squared reconstruction loss, and the derivative is computed by automatic differentiation; the penalty points are drawn by Latin hypercube sampling across the spatial and parameter domain because the smoothness constraint must hold everywhere, not just at supervised points. In the paper's formulation, this enforces the limit $\lim_{\varphi_1\to\varphi_2}|F(X,\varphi_1)-F(X,\varphi_2)|\approx 0$, i.e., Lipschitz continuity along the flow-rate axis. The same field is then differentiated against a per-layer objective, $\varphi(z)=\arg\min_\varphi |F_\Theta(x,y|z)-\mathrm{layer}(z)|$, to choose optimal flow rates for each printed layer.

What would settle it

Print several copies of one geometry at each of several densely spaced flow rates (for example, five copies at each 10% step from 50% to 250%), CT-scan them, and compare the GDIR-predicted volumes against these ground truths; if the GDIR errors are systematically largest exactly in regions where volume changes sharply with flow, such as thin-wall formation or the bunny ears, the smoothness penalty has imposed wrong structure. A cheaper version is to measure physical mass at those rates and test whether the nearly linear digital-weight curve from Figure 4 matches the true mass-versus-flow relationship within measurement error.

Watch

Extended reading notes

Core claim

The paper's central claim is that penalizing $\|\partial F_\Theta/\partial \varphi\|_2^2$ at randomly sampled points in the combined spatial-flow domain forces the occupancy or signed-distance field $F_\Theta(x,y,z;\varphi)$ to interpolate smoothly between the discrete flow-rate conditions seen in training. Trained jointly with the usual reconstruction loss, this penalty yields Lipschitz-like continuity along the flow-rate axis, so the network can 'imagine' the geometry that would result from an unseen flow rate instead of collapsing or oscillating in sparsely supervised regions. The authors support the claim with quantitative comparisons at held-out flow rates (250% for the bolt, 115% for the bunny, 70% for the gear, 35% for the statue), where GDIR networks outperform unregularized SIRENs on L1 and SSIM metrics, and with a physical-versus-digital weight comparison showing a nearly linear, physically intuitive relationship between flow rate and reconstructed mass. They then use the field as a simulator to select, for each of the bunny's 100 layers, the flow rate that minimizes the L1 difference between the reconstructed cross-section and the expected-geometry cross-section, and report that a bunny printed with these per-layer rates has smoother contours and better ear detail.

Load-bearing premise

The load-bearing premise is that pushing the field's output to change only slowly with flow rate, at randomly sampled points, makes its predictions more physically correct rather than artificially flattening real geometry–flow relationships.

Editorial extensions

If this is right

  • With GDIR, a neural field trained on a handful of printed flow-rate settings can predict produced geometry at unseen settings, so geometric fidelity can be evaluated without manufacturing a test part.
  • Per-layer flow-rate optimization can improve geometric fidelity in feature-rich parts; in the bunny demo, ears printed better at higher flow while the body benefited from slight under-extrusion.
  • Because the method is process- and representation-agnostic, the same regularization could be applied to other manufacturing parameters whose variation changes the output geometry continuously.
  • Occupancy and signed-distance variants both benefit from GDIR at unseen flow rates, and the SDF variant generally gives the best unseen-condition accuracy in the paper's comparisons.
  • The regularization slows training by about 8% but adds no inference cost, because gradients with respect to $\varphi$ are not needed when evaluating the trained field.

Reading between the lines

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

  • Beyond the paper, GDIR's success suggests the same derivative-penalty idea could regularize interpolation along other process variables—temperature, print speed, or layer height—whenever the variable is known to affect geometry continuously, turning a small set of prints into a continuous process-response model.
  • The nearly linear digital-weight curve should be tested against a denser set of physical prints: if the true mass-versus-flow relationship has measurable curvature, the penalty that flattens SIREN's oscillations could also be flattening genuine nonlinearity.
  • A natural stress test is to apply GDIR to a geometry with a sharp, threshold-like response to flow, such as a thin wall that only forms above a critical rate; if the penalty still produces smooth interpolation there, it would confirm that GDIR follows the data where supervised rather than imposing smoothness everywhere.
  • The per-layer optimization treats each layer independently; an obvious extension is to add a coupling term between neighboring layers so that support and adhesion effects, like the rough belly noted in the paper, enter the objective.
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 / 6 minor

Summary. The paper proposes a 4D neural-field representation of 3D printed parts, with inputs (x, y, z, flow rate φ), trained on CT scans of four geometries each printed at nine flow rates. To improve interpolation at unseen flow rates, it adds a 'gradient-driven interpolation regularization' (GDIR) term that penalizes the squared norm of ∂F/∂φ. The authors report that GDIR variants outperform plain SIREN at held-out flow rates (Table A2) and use the trained field to optimize per-layer flow rates for a bunny model, with a qualitative print-quality improvement.

Significance. If the method were fully validated, it would be a practically useful contribution to data-driven 3D printing process optimization, and the open dataset and code are valuable assets. The idea of treating a process parameter as a continuous dimension of a neural field is appealing and clearly presented. However, the theoretical justification of the regularization is flawed and the quantitative evidence is thin: the central claim of physically correct smooth interpolation is not yet established, and the optimization demonstration is qualitative on a single geometry.

major comments (4)
  1. [§4.2.2, Eqs. (5)–(7)] The derivation is invalid: Eq. (6), ∂FΘ/∂φ = 0, is far stronger than the Lipschitz condition in Eq. (5), which only bounds |F(φ1) − F(φ2)| by c|φ1 − φ2|. A Lipschitz function can have a nonzero derivative, whereas the GDIR penalty actively drives the derivative to zero, i.e., it is a zero-slope prior rather than a Lipschitz regularizer. Because the true field changes strongly with flow rate (as shown qualitatively in Fig. 1C and quantitatively by the weight-vs-flow trends in Fig. 4), this regularizer fights the signal it is intended to interpolate. The statement that GDIR 'encourages Lipschitz continuity' is therefore unsupported. The authors should either reformulate the theoretical motivation or implement an actual Lipschitz constraint; alternatively, they should explicitly characterize the penalty as a zero-derivative smoothing prior and justify why flattening is desirable in the gaps between observed flow rates.
  2. [§2.2, Table A2] The generalization evidence rests on a single held-out print per geometry: Bolt at 250%, Bunny at 115%, Gear at 70%, and Statue at 35%. Table A2 reports no error bars for L1 and only slice-level SSIM variability, so each unseen-condition comparison has n = 1 and no statistical significance can be assessed. Moreover, no non-neural baseline (e.g., linear interpolation, nearest-neighbor, or Gaussian-process interpolation of the nine registered volumes) is provided. The observed improvements over SIREN could reflect the regularizer's smoothing effect rather than physically accurate prediction. At minimum, the authors should provide multiple prints per unseen flow rate and compare against a standard volumetric interpolant.
  3. [Fig. 4 and §2.2] The 'almost linear' digital-weight-vs-flow-rate plot is not independent validation of the method. This curve is generated by the very GDIR field whose regularization explicitly penalizes ∂F/∂φ, so a nearly flat or linear trend is a direct and expected consequence of the regularizer; it shows that the regularizer works as a smoothness penalty, not that the predicted trend is physically correct. To make this a meaningful validation, the authors should compare the GDIR-predicted digital weights at the four held-out flow rates with the measured physical weights of the corresponding held-out prints, or use another metric evaluated on independent measurements.
  4. [§2.3 and §4.4] The optimization demonstration is qualitative: it covers one geometry (Bunny), uses one printed realization, and reports no quantitative fidelity metrics (e.g., L1 or SSIM against the expected CAD/slicer geometry) and no comparison to a baseline such as the standard 100% flow print or an unregularized SIREN-based optimization. Since the optimizer in Eq. (8) treats FΘ as a simulator, any systematic flattening artifact in the field will directly bias the selected per-layer flow rates. The paper should report quantitative geometry-comparison metrics for the optimized print and include at least one control condition to support the claim of improved geometric fidelity.
minor comments (6)
  1. [§2.2, Eq. (1)] Equation (1) is not a definition of Lipschitz continuity; it is a pointwise continuity statement. The notation is also malformed ('∀X ∈ R3 lim...' should be written as a limit with an explicit inequality). The authors should use Eq. (5) as the formal definition and clarify the relationship.
  2. [§4.2.1, Eqs. (2) and (3)] Equations (2) and (3) are duplicates; one should be removed or the surrounding text should be corrected.
  3. [§4.2.2, Eq. (6)] The notation '∂FΘ/∂φ = 0' is inconsistent with the squared-norm loss in Eq. (7); the equation should state that the squared norm of the partial derivative is penalized.
  4. [Table A2] The held-out flow rates (Bolt 250%, Bunny 115%, Gear 70%, Statue 35%) are given only in the main text; they should appear in the table headers or caption for readability.
  5. [Fig. 4] The legend 'Ground truth weight and CT scan weight' is ambiguous; the authors should distinguish clearly among physical scale weight, digital weight from CT-derived occupancy, and digital weight from the neural fields.
  6. [§4.3.1] The grid search for the sinusoidal frequency is performed on 10% of the training data points; please state explicitly that the held-out flow-rate prints were not used in this hyperparameter selection, to rule out information leakage.

Circularity Check

1 steps flagged · score 2.0 of 10

Core interpolation claim survives external unseen-flow-rate CT tests; only the Figure 4 weight-trend validation restates the smoothness penalty.

  1. other [Section 2.2, Figure 4; Methods 4.2.2, Eq. (7)]
    "The physical to digital weight comparison serves as another quantitative indicator which shows the stability of the GDIR networks in predicting the volumetric structure of a geometry at any flow rate value... While the GDIR method outputs an almost linear relationship between material flow rate and sample weight, SIREN exhibits a wave-like pattern, particularly in the over-extrusion regime with sparse supervision."

    GDIR is implemented as L = L_MSE + lambda * ||dF_Theta/d_phi||_2^2 (Eq. 7). Penalizing the phi-derivative at all sampled coordinates directly biases F to change slowly with phi, so reconstructed occupancy, and therefore digital weight versus phi, is expected to be smooth and nearly monotone or linear. Using this 'almost linear' trend as a quantitative indicator that the network correctly captures the geometry-flow relationship is therefore partly a restatement of the regularizer rather than independent evidence. This does not affect the main unseen-flow-rate comparison in Table A2, which uses held-out CT scans.

full rationale

The paper's central claim, that GDIR enables 4D neural fields to predict geometry at unseen flow rates and that these fields can optimize per-layer flow rates, is supported by comparisons against withheld CT scans at 250% (Bolt), 115% (Bunny), 70% (Gear), and 35% (Statue). These are external benchmarks not used in fitting; the only tuned hyperparameters (lambda and sinusoidal frequency) were selected on training or held-in data and do not directly force the reported improvements on unseen prints. No load-bearing self-citation chain is present; prior works by the same group (refs. 5, 6, 7, 15, 16) are background or application references, not the source of GDIR's validity. The one partial circularity is the Figure 4 weight-versus-flow validation, which is a consequence of the same smoothness penalty being validated, so it does not independently confirm physical correctness. The mathematical slippage from Lipschitz continuity (Eq. 5) to the zero-derivative penalty (Eq. 6) is a correctness or implementation concern, not circularity, because the regularizer is explicitly defined by the zero-derivative target. The unseen-condition evaluation rests on a single test print per geometry, which is a statistical-thinness concern rather than a circularity concern. Overall, the core derivation is self-contained against external benchmarks, with only a minor validation circularity.

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

The central method relies on one ad-hoc regularization assumption (zero partial derivative) plus standard domain assumptions about print physics and CT fidelity. No new physical entities are introduced. The only fitted hyperparameters are lambda and the sinusoidal frequency omega.

free parameters (2)
  • GDIR weight lambda = empirically tuned
    Balances the MSE and gradient penalty in Equation 7; tuned per geometry to balance smoothness and fidelity.
  • SIREN activation frequency omega = not reported
    Found via grid search on 10% of the available training data (Methods 4.3.1).
assumptions (4)
  • domain assumption Geometry varies smoothly with flow rate (small-motion assumption)
    Used to justify GDIR (Section 2.2, Methods 4.2.2); assumes small changes in flow rate produce small geometry changes, which may fail in feature-sensitive regions such as the bunny ears.
  • domain assumption Digital weight should be proportional to flow rate
    Used to interpret Figure 4 as validation; assumes a linear mass-flow relationship, which is an approximation and not independently established.
  • domain assumption CT scans and registration provide accurate ground truth geometry
    All training labels come from micro-CT scans with thresholding and CPD registration (Methods 4.1.2, 4.1.3); errors here propagate directly to the learned field.
  • ad hoc to paper Minimizing the partial derivative with respect to flow rate at random points enforces the desired Lipschitz-like continuity
    The regularization in Equations 6-7 is a modeling choice that approximates continuity but is stronger (forces flatness) and is not derived from physics; this is the key ad hoc element.

how reviews work

0 comments
Cite this review

Pith. "Pith review of Regularized interpolation in 4D neural fields enables optimization of 3D printed geometries." pith.science (2026). https://pith.science/paper/CI6H5NTN

@misc{pith2026250201517,
  author       = {Pith},
  title        = {Pith review of: Regularized interpolation in 4D neural fields enables optimization of 3D printed geometries},
  year         = {2026},
  howpublished = {\url{https://pith.science/paper/CI6H5NTN}},
  note         = {Machine review of arXiv:2502.01517}
}
read the original abstract

The ability to accurately produce geometries with specified properties is perhaps the most important characteristic of a manufacturing process. 3D printing is marked by exceptional design freedom and complexity but is also prone to geometric and other defects that must be resolved for it to reach its full potential. Ultimately, this will require both astute design decisions and timely parameter adjustments to maintain stability that is challenging even with expert human operators. While machine learning is widely investigated in 3D printing, existing methods typically overlook spatial features that vary across prints and thus find it difficult to produce desired geometries. Here, we encode volumetric representations of printed parts into neural fields and apply a new regularization strategy, based on minimizing the partial derivative of the field's output with respect to a single, non-learnable parameter. By thus encouraging small input changes to yield only small output variations, we encourage smooth interpolation between observed volumes and hence realistic geometry predictions. This framework therefore allows the extraction of 'imagined' 3D shapes, revealing how a part would look if manufactured under previously unseen parameters. The resulting continuous field is used for data-driven optimization to maximize geometric fidelity between expected and produced geometries, reducing post-processing, material waste, and production costs. By optimizing process parameters dynamically, our approach enables advanced planning strategies, potentially allowing manufacturers to better realize complex and feature-rich designs.

Figures

Figures reproduced from arXiv: 2502.01517 by the authors.

Figure 1
Figure 1. Data collection and processing pipeline. A. Sample axial and voxel frames obtained from the computer tomography procedure. B. Data processing pipeline developed for cleaning and registering different trials from the same geometry manufactured at different flow rate values. C. Rendered parts from the Bolt trials. initial layers to improve adhesion between the build surface and the part, while lower flow rates help ma… view at source ↗
Figure 2
Figure 2. End-to-end domain solver. The domain is solved for a specific flow rate value, ϕ, through the trained 4D neural field. The calculation of the partial derivative of the output fΘ with respect to ϕ and minimization of its magnitude enables smooth interpolation between seen ϕs. 2 Results 2.1 Dataset and problem formulation Process parameters in manufacturing, like material flow rate in 3D printing, are critical in alig… view at source ↗
Figure 3
Figure 3. Interpolation results. While SIREN (blue) collapse at unseen regimes, our network (pink) generates smoother interpolation results [PITH_FULL_IMAGE:figures/full_fig_p004_3.png] view at source ↗
Figures from the paper (2 more)
Figure 4
Figure 4. Figure 4: Physical vs Digital weights. We compare the physical and digital weights of the samples, reconstructed from CT scans, SIREN and our networks. and Si ∈ 0, 1 the material state. The number of data points N, is given by the product of the dimensions of the 3D array repres…
Figure 5
Figure 5. Figure 5: Geometric fidelity optimization results. A. Renders of the produced geometry of the Stanford bunny as manufactured at 100% and 170% flow rate values. B. The optimization pipeline compares the learned field reconstructions with renderings from the expected geometry. C. …

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. Hybrid Reasoning for Perception, Explanation, and Autonomous Action in Manufacturing

    cs.AI 2025-06 conditional novelty 6.0 of 10

    CIPHER embeds a convolutional regression expert into a vision-language-action model, enabling a commercial 3D printer to perceive extrusion state, reason about faults, and generate corrective G-code from images or text.

Reference graph

Works this paper leans on

67 extracted references · 65 canonical work pages · cited by 1 Pith paper

  1. [1]

    Material extrusion additive manufacturing of multifunctional sandwich panels with load-bearing and acoustic capabilities for aerospace applications

    Juliette Pierre, Filippo Iervolino, Rouhollah D Farahani, Nicola Piccirelli, Martin Lévesque, and Daniel Therriault. Material extrusion additive manufacturing of multifunctional sandwich panels with load-bearing and acoustic capabilities for aerospace applications. Additive Manufacturing, 61:103344, 2023

    work page 2023

  2. [2]

    Review of additive manufacturing technologies and applications in the aerospace industry, pages 7–31

    Joel C Najmon, Sajjad Raeisi, and Andres Tovar. Review of additive manufacturing technologies and applications in the aerospace industry, pages 7–31. Elsevier, 2019

    work page 2019

  3. [3]

    3d printed different polymer fuel grains for hybrid rocket engine

    Xiaodong Yu, Hongsheng Yu, Hongwei Gao, Wei Zhang, Luigi T DeLuca, and Ruiqi Shen. 3d printed different polymer fuel grains for hybrid rocket engine. FirePhysChem, 2023

    work page 2023

  4. [4]

    Fused deposition modeling for unmanned aerial vehicles (uavs): A review

    Helge Klippstein, Alejandro Sanchez, Hany Hassanin, Yahya Zweiri, and Lakmal Seneviratne. Fused deposition modeling for unmanned aerial vehicles (uavs): A review. Advanced Engineering Materials, 20:1700552, 11 2017

    work page 2017

  5. [5]

    Facile method for 3d printing conformally onto uneven surfaces and its application to face masks

    Douglas A J., Samson Kerr D G., Pattinson Sebastian W Ji Zehao, and Brion. Facile method for 3d printing conformally onto uneven surfaces and its application to face masks. Scientific Reports, 13:21659, 12 2023

    work page 2023

  6. [6]

    Additive manufacturing of cellulosic materials with robust mechanics and antimicrobial functionality

    Sebastian W Pattinson and A John Hart. Additive manufacturing of cellulosic materials with robust mechanics and antimicrobial functionality. Advanced Materials Technologies, 2:1600084, 2017

    work page 2017

  7. [7]

    Additive manufacturing of biomechanically tailored meshes for compliant wearable and implantable devices

    Sebastian W Pattinson, Meghan E Huber, Sanha Kim, Jongwoo Lee, Sarah Grunsfeld, Ricardo Roberts, Gregory Dreifus, Christoph Meier, Lei Liu, Neville Hogan, and A John Hart. Additive manufacturing of biomechanically tailored meshes for compliant wearable and implantable devices. Advanced Functional Materials, 29:1901815, 2019

    work page 2019

  8. [8]

    Reconfigurable aqueous 3d printing with adaptive dual locks

    Yuchen Fu, Zhiyong Li, Sai Zhao, Honghao Hou, and Yu Chai. Reconfigurable aqueous 3d printing with adaptive dual locks. Science Advances, 10:eadk4080, 2024

    work page 2024

Show all 67 references
  1. [9]

    3d printed patient-specific aortic root models with internal sensors for minimally invasive applications

    Ghazaleh Haghiashtiani, Kaiyan Qiu, Jorge D Zhingre Sanchez, Zachary J Fuenning, Priya Nair, Sarah E Ahlberg, Paul A Iaizzo, and Michael C McAlpine. 3d printed patient-specific aortic root models with internal sensors for minimally invasive applications. Science Advances, 6:ea...

    2020

  2. [10]

    3d printing of buildings and building components as the future of sustainable construction? Procedia Engineering, 151:292–299, 2016

    Izabela Hager, Anna Golonka, and Roman Putanowicz. 3d printing of buildings and building components as the future of sustainable construction? Procedia Engineering, 151:292–299, 2016. Ecology and new building materials and products 2016

    2016

  3. [11]

    NASA Centennial Challenge: Three Dimensional (3D) Printed Habitat, pages 333–342

    Monsi C Roman, Eric A Eberly, Robert P Mueller, and Scott Deutsch. NASA Centennial Challenge: Three Dimensional (3D) Printed Habitat, pages 333–342

  4. [12]

    Additive manufacturing (3d printing): A review of materials, methods, applications and challenges

    Tuan D Ngo, Alireza Kashani, Gabriele Imbalzano, Kate T Q Nguyen, and David Hui. Additive manufacturing (3d printing): A review of materials, methods, applications and challenges. Composites Part B: Engineering, 143:172–196, 2018

    2018

  5. [13]

    Fused deposition modeling of thermoplastic elastomeric materials: Challenges and opportunities

    Pratiksha Awasthi and Shib Shankar Banerjee. Fused deposition modeling of thermoplastic elastomeric materials: Challenges and opportunities. Additive Manufacturing, 46, 2021. Cited by: 109

    2021

  6. [14]

    Optimization of fused deposition modeling process parameters: a review of current research and future prospects

    Syed H., Bhowmik Jahar L Mohamed Omar A., and Masood. Optimization of fused deposition modeling process parameters: a review of current research and future prospects. Advances in Manufacturing, 3:42–53, 3 2015

    2015

  7. [15]

    Generalisable 3d printing error detection and correction via multi-head neural networks

    Douglas A J Brion and Sebastian W Pattinson. Generalisable 3d printing error detection and correction via multi-head neural networks. Nature Communications, 13:4654, 2022. 10 Margadji et al

    2022

  8. [16]

    Quantitative and real-time control of 3d printing material flow through deep learning

    Douglas A J Brion and Sebastian W Pattinson. Quantitative and real-time control of 3d printing material flow through deep learning. Advanced Intelligent Systems, 4:2200153, 2022

    2022

  9. [17]

    Image analysis-based closed loop quality control for additive manufacturing with fused filament fabrication

    Chenang Liu, Andrew Chung Chee Law, David Roberson, and Zhenyu (James) Kong. Image analysis-based closed loop quality control for additive manufacturing with fused filament fabrication. Journal of Manufacturing Systems, 51:75–86, 2019

    2019

  10. [18]

    Autonomous in-situ correction of fused deposition modeling printers using computer vision and deep learning

    Zeqing Jin, Zhizhou Zhang, and Grace X Gu. Autonomous in-situ correction of fused deposition modeling printers using computer vision and deep learning. Manufacturing Letters, 22:11–15, 2019

    2019

  11. [19]

    Design for additive manufacturing: Trends, opportunities, considerations, and constraints

    Mary Kathryn Thompson, Giovanni Moroni, Tom Vaneker, Georges Fadel, R Ian Campbell, Ian Gibson, Alain Bernard, Joachim Schulz, Patricia Graf, Bhrigu Ahuja, and Filomeno Martina. Design for additive manufacturing: Trends, opportunities, considerations, and constraints. CIRP Ann...

    2016

  12. [20]

    3d masked autoencoders with application to anomaly detection in non-contrast enhanced breast mri

    Daniel M Lang, Eli Schwartz, Cosmin I Bercea, Raja Giryes, and Julia A Schnabel. 3d masked autoencoders with application to anomaly detection in non-contrast enhanced breast mri. arXiv preprint arXiv:2303.05861, 2023

    2023 arXiv

  13. [21]

    Gaussian splatting slam

    Hidenobu Matsuki, Riku Murai, Paul H J Kelly, and Andrew J Davison. Gaussian splatting slam. Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition, 2024

    2024

  14. [22]

    S3cnet: A sparse semantic scene completion network for lidar point clouds

    Ran Cheng, Christopher Agia, Yuan Ren, Xinhai Li, and Liu Bingbing. S3cnet: A sparse semantic scene completion network for lidar point clouds. In Jens Kober, Fabio Ramos, and Claire Tomlin, editors, Proceedings of the 2020 Conference on Robot Learning, volume 155, pages 2148–2...

    2020

  15. [23]

    Geometric deep learning: Grids, groups, graphs, geodesics, and gauges, 2021

    Michael M Bronstein, Joan Bruna, Taco Cohen, and Petar Veliˇckovi´c. Geometric deep learning: Grids, groups, graphs, geodesics, and gauges, 2021

    2021

  16. [24]

    Graph neural network for traffic forecasting: A survey

    Weiwei Jiang and Jiayun Luo. Graph neural network for traffic forecasting: A survey. Expert Systems with Applications, 207:117921, 11 2022

    2022

  17. [25]

    Stretchable e-skin and transformer enable high-resolution morphological reconstruction for soft robots

    Francesco, Zhang Shiming, Yang Yunjie Hu Delin, and Giorgio-Serchi. Stretchable e-skin and transformer enable high-resolution morphological reconstruction for soft robots. Nature Machine Intelligence, 5:261–272, 3 2023

    2023

  18. [26]

    Geometric deep learning for shape correspondence in mass customization by three-dimensional printing

    Jida Huang, Hongyue Sun, Tsz-Ho Kwok, Chi Zhou, and Wenyao Xu. Geometric deep learning for shape correspondence in mass customization by three-dimensional printing. Journal of Manufacturing Science and Engineering, 142:61003, 11 2020

    2020

  19. [27]

    Learning mesh-based simulation with graph networks, 2021

    Tobias Pfaff, Meire Fortunato, Alvaro Sanchez-Gonzalez, and Peter W Battaglia. Learning mesh-based simulation with graph networks, 2021

    2021

  20. [28]

    3d representation methods: A survey, 2024

    Zhengren Wang. 3d representation methods: A survey, 2024

    2024

  21. [29]

    Joint implicit neural representation for high-fidelity and compact vector fonts

    Chia-Hao Chen, Ying-Tian Liu, Zhifei Zhang, Yuan-Chen Guo, and Song-Hai Zhang. Joint implicit neural representation for high-fidelity and compact vector fonts. In 2023 IEEE/CVF International Conference on Computer Vision (ICCV), pages 5515–5525, 2023

    2023

  22. [30]

    Neural geometric level of detail: Real-time rendering with implicit 3d shapes, 2021

    Towaki Takikawa, Joey Litalien, Kangxue Yin, Karsten Kreis, Charles Loop, Derek Nowrouzezahrai, Alec Jacobson, Morgan McGuire, and Sanja Fidler. Neural geometric level of detail: Real-time rendering with implicit 3d shapes, 2021

    2021

  23. [31]

    Learning continuous image representation with local implicit image function, 2021

    Yinbo Chen, Sifei Liu, and Xiaolong Wang. Learning continuous image representation with local implicit image function, 2021

    2021

  24. [32]

    A multi-implicit neural representation for fonts, 2022

    Pradyumna Reddy, Zhifei Zhang, Matthew Fisher, Hailin Jin, Zhaowen Wang, and Niloy J Mitra. A multi-implicit neural representation for fonts, 2022

    2022

  25. [33]

    Exploring the performance of implicit neural representations for brain image registration

    Charissa, Rachmadi Muhammad Febrian, Schlachter Matthias, Skibbe Henrik Byra Michal, and Poon. Exploring the performance of implicit neural representations for brain image registration. Scientific Reports, 13:17334, 10 2023

    2023

  26. [34]

    Freenerf: Improving few-shot neural rendering with free frequency regularization, 2023

    Jiawei Yang, Marco Pavone, and Yue Wang. Freenerf: Improving few-shot neural rendering with free frequency regularization, 2023

    2023

  27. [35]

    Bundlesdf: Neural 6-dof tracking and 3d reconstruction of unknown objects, 2023

    Bowen Wen, Jonathan Tremblay, Valts Blukis, Stephen Tyree, Thomas Muller, Alex Evans, Dieter Fox, Jan Kautz, and Stan Birchfield. Bundlesdf: Neural 6-dof tracking and 3d reconstruction of unknown objects, 2023

    2023

  28. [36]

    isdf: Real-time neural signed distance fields for robot perception, 2022

    Joseph Ortiz, Alexander Clegg, Jing Dong, Edgar Sucar, David Novotny, Michael Zollhoefer, and Mustafa Mukadam. isdf: Real-time neural signed distance fields for robot perception, 2022

    2022

  29. [37]

    Shapo: Implicit representations for multi-object shape, appearance, and pose optimization, 2022

    Muhammad Zubair Irshad, Sergey Zakharov, Rares Ambrus, Thomas Kollar, Zsolt Kira, and Adrien Gaidon. Shapo: Implicit representations for multi-object shape, appearance, and pose optimization, 2022

    2022

  30. [38]

    ilabel: Revealing objects in neural fields

    Shuaifeng Zhi, Edgar Sucar, Andre Mouton, Iain Haughton, Tristan Laidlow, and Andrew J Davison. ilabel: Revealing objects in neural fields. IEEE Robotics and Automation Letters, 8:832–839, 2023. 11 Margadji et al

    2023

  31. [39]

    Feature-realistic neural fusion for real-time, open set scene understanding

    Kirill Mazur, Edgar Sucar, and Andrew J Davison. Feature-realistic neural fusion for real-time, open set scene understanding. In 2023 IEEE International Conference on Robotics and Automation (ICRA), pages 8201–8207, 2023

    2023

  32. [40]

    Physics informed deep learning for computational elastodynamics without labeled data, 2020

    Chengping Rao, Hao Sun, and Yang Liu. Physics informed deep learning for computational elastodynamics without labeled data, 2020

    2020

  33. [41]

    Nsfnets (navier-stokes flow nets): Physics- informed neural networks for the incompressible navier-stokes equations

    Xiaowei Jin, Shengze Cai, Hui Li, and George Em Karniadakis. Nsfnets (navier-stokes flow nets): Physics- informed neural networks for the incompressible navier-stokes equations. Journal of Computational Physics, 426:109951, 2 2021

    2021

  34. [42]

    Physics-informed neural networks: A deep learning framework for solving forward and inverse problems involving nonlinear partial differential equations

    M Raissi, P Perdikaris, and G E Karniadakis. Physics-informed neural networks: A deep learning framework for solving forward and inverse problems involving nonlinear partial differential equations. Journal of Computational Physics, 378:686–707, 2019

    2019

  35. [43]

    Implicit neural representations with periodic activation functions, 2020

    Vincent Sitzmann, Julien N P Martel, Alexander W Bergman, David B Lindell, and Gordon Wetzstein. Implicit neural representations with periodic activation functions, 2020

    2020

  36. [44]

    Compositional pattern producing networks : A novel abstraction of development

    Kenneth O Stanley. Compositional pattern producing networks : A novel abstraction of development. 2007

    2007

  37. [45]

    Inras: Implicit neural representation for audio scenes

    Kun Su, Mingfei Chen, and Eli Shlizerman. Inras: Implicit neural representation for audio scenes. In Alice H Oh, Alekh Agarwal, Danielle Belgrave, and Kyunghyun Cho, editors, Advances in Neural Information Processing Systems, 2022

    2022

  38. [46]

    Hypersound: Generat- ing implicit neural representations of audio signals with hypernetworks, 2024

    Filip Szatkowski, Karol J Piczak, Przemysław Spurek, Jacek Tabor, and Tomasz Trzci´nski. Hypersound: Generat- ing implicit neural representations of audio signals with hypernetworks, 2024

    2024

  39. [47]

    Meshfreeflownet: A physics-constrained deep continuous space-time super-resolution framework, 2020

    Chiyu Max Jiang, Soheil Esmaeilzadeh, Kamyar Azizzadenesheli, Karthik Kashinath, Mustafa Mustafa, Hamdi A Tchelepi, Philip Marcus, Prabhat, and Anima Anandkumar. Meshfreeflownet: A physics-constrained deep continuous space-time super-resolution framework, 2020

    2020

  40. [48]

    Implicit geometric regularization for learning shapes, 2020

    Amos Gropp, Lior Yariv, Niv Haim, Matan Atzmon, and Yaron Lipman. Implicit geometric regularization for learning shapes, 2020

    2020

  41. [49]

    Neural descriptor fields: Se(3)-equivariant object representations for manipulation, 2021

    Anthony Simeonov, Yilun Du, Andrea Tagliasacchi, Joshua B Tenenbaum, Alberto Rodriguez, Pulkit Agrawal, and Vincent Sitzmann. Neural descriptor fields: Se(3)-equivariant object representations for manipulation, 2021

    2021

  42. [50]

    Capturing dynamical correlations using implicit neural representations

    Sathya R Chitturi, Zhurun Ji, Alexander N Petsch, Cheng Peng, Zhantao Chen, Rajan Plumley, Mike Dunne, Sougata Mardanya, Sugata Chowdhury, Hongwei Chen, Arun Bansil, Adrian Feiguin, Alexander I Kolesnikov, Dharmalingam Prabhakaran, Stephen M Hayden, Daniel Ratner, Chunjing Jia...

    2023

  43. [51]

    Occupancy networks: Learning 3d reconstruction in function space, 2019

    Lars Mescheder, Michael Oechsle, Michael Niemeyer, Sebastian Nowozin, and Andreas Geiger. Occupancy networks: Learning 3d reconstruction in function space, 2019

    2019

  44. [52]

    Deepsdf: Learning continuous signed distance functions for shape representation, 2019

    Jeong Joon Park, Peter Florence, Julian Straub, Richard Newcombe, and Steven Lovegrove. Deepsdf: Learning continuous signed distance functions for shape representation, 2019

    2019

  45. [53]

    Nerf: Representing scenes as neural radiance fields for view synthesis, 2020

    Ben Mildenhall, Pratul P Srinivasan, Matthew Tancik, Jonathan T Barron, Ravi Ramamoorthi, and Ren Ng. Nerf: Representing scenes as neural radiance fields for view synthesis, 2020

    2020

  46. [54]

    Fourier features let networks learn high frequency functions in low dimensional domains, 2020

    Matthew Tancik, Pratul P Srinivasan, Ben Mildenhall, Sara Fridovich-Keil, Nithin Raghavan, Utkarsh Singhal, Ravi Ramamoorthi, Jonathan T Barron, and Ren Ng. Fourier features let networks learn high frequency functions in low dimensional domains, 2020

    2020

  47. [55]

    Plenoctrees for real-time rendering of neural radiance fields, 2021

    Alex Yu, Ruilong Li, Matthew Tancik, Hao Li, Ren Ng, and Angjoo Kanazawa. Plenoctrees for real-time rendering of neural radiance fields, 2021

    2021

  48. [56]

    Acorn: Adaptive coordinate networks for neural scene representation, 2021

    Julien N P Martel, David B Lindell, Connor Z Lin, Eric R Chan, Marco Monteiro, and Gordon Wetzstein. Acorn: Adaptive coordinate networks for neural scene representation, 2021

    2021

  49. [57]

    Instant neural graphics primitives with a multiresolution hash encoding

    Thomas Müller, Alex Evans, Christoph Schied, and Alexander Keller. Instant neural graphics primitives with a multiresolution hash encoding. ACM Transactions on Graphics, 41:1–15, 7 2022

    2022

  50. [58]

    Neural fields in visual computing and beyond

    Yiheng Xie, Towaki Takikawa, Shunsuke Saito, Or Litany, Shiqin Yan, Numair Khan, Federico Tombari, James Tompkin, Vincent Sitzmann, and Srinath Sridhar. Neural fields in visual computing and beyond. Computer Graphics Forum, 2022

    2022

  51. [59]

    Learning smooth neural functions via lipschitz regularization

    Hsueh-Ti Derek Liu, Francis Williams, Alec Jacobson, Sanja Fidler, and Or Litany. Learning smooth neural functions via lipschitz regularization. ACM SIGGRAPH 2022 Conference Proceedings, 2022. 12 Margadji et al

    2022

  52. [60]

    Double backpropagation increasing generalization performance

    Harris Drucker and Yann Le Cun. Double backpropagation increasing generalization performance. IJCNN-91- Seattle International Joint Conference on Neural Networks, ii:145–150 vol.2, 1991

    1991

  53. [61]

    Robustness via curvature regularization, and vice versa

    Seyed-Mohsen Moosavi-Dezfooli, Alhussein Fawzi, Jonathan Uesato, and Pascal Frossard. Robustness via curvature regularization, and vice versa. 2019 IEEE/CVF Conference on Computer Vision and Pattern Recognition (CVPR), pages 9070–9078, 2018

    2019

  54. [62]

    Intuitive shape editing in latent space

    Tim Elsner, Moritz Ibing, Victor Czech, Julius Nehring-Wirxel, and Leif Kobbelt. Intuitive shape editing in latent space. ArXiv, abs/2111.12488, 2021

    2021 arXiv

  55. [63]

    Frame interpolation in video stream using optical flow methods

    S Lyasheva, R Rakhmankulov, and M Shleymovich. Frame interpolation in video stream using optical flow methods. Journal of Physics: Conference Series, 1488:12024, 3 2020

    2020

  56. [64]

    Automatic differentiation in pytorch

    Adam Paszke, Sam Gross, Soumith Chintala, Gregory Chanan, Edward Yang, Zachary DeVito, Zeming Lin, Alban Desmaison, Luca Antiga, and Adam Lerer. Automatic differentiation in pytorch. 2017

    2017

  57. [65]

    Marching cubes: A high resolution 3d surface construction algorithm

    William E Lorensen and Harvey E Cline. Marching cubes: A high resolution 3d surface construction algorithm. In Proceedings of the 14th Annual Conference on Computer Graphics and Interactive Techniques, pages 163–169. Association for Computing Machinery, 1987

    1987

  58. [66]

    Ntopo: Mesh-free topology optimization using implicit neural representations, 2021

    Jonas Zehnder, Yue Li, Stelian Coros, and Bernhard Thomaszewski. Ntopo: Mesh-free topology optimization using implicit neural representations, 2021

    2021

  59. [67]

    Nito: Neural implicit fields for resolution-free topology optimization, 2024

    Amin Heyrani Nobari, Giorgio Giannone, Lyle Regenwetter, and Faez Ahmed. Nito: Neural implicit fields for resolution-free topology optimization, 2024. 13 Margadji et al. Appendix This document includes appendix tables and figures. 14 Margadji et al. Table A1: Quantitative comp...

    2024

Pith tools

Reviewed August 9, 2026 · model on record in the stance chip above.