Pith. sign in

REVIEW 4 major objections 6 minor 73 references

Visualizing the Invisible: A Generative AR System for Intuitive Multi-Modal Sensor Data Presentation

T0 review · 4 major / 6 minor · reviewed 2026-08-11 · deepseek-v4-flash

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

Pith's one-line read Vivar maps multi-modal sensor data into the CLIP embedding space via barycentric interpolation between anchor readings, so that diffusion models generate AR visualizations that accurately and consistently reflect sensor values, with…

desk verdict A solid end-to-end AR sensor visualization system with a real user study, but the central 'accuracy' claim leans on by-construction smoothness and perceived quality rather than objective validation. read the letter →

arxiv 2412.13509 v2 pith:DBDXXTZE submitted 2024-12-18 cs.HC

Yunqi Guo , Kaiyuan Hou , Heming Fu , Hongkai Chen , Zhenyu Yan , Guoliang Xing , Xiaofan Jiang This is my paper · ORCID
classification cs.HC
keywords augmentedrealitysensordatavisualizationcross-modalembeddingbarycentricinterpolationCLIP3DGaussianSplattinglatentreusegenerativeAI
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

Vivar is an end-to-end system that turns raw multi-modal sensor readings—air quality, temperature, humidity, emotion, sound—into 3D augmented-reality scenes that a non-expert can interpret without training. Its core claim is that sensor values can be embedded into the CLIP image-text embedding space by barycentric interpolation between anchor readings, and that this embedding, when fed to a diffusion model, produces visualizations that accurately reflect value changes and stay coherent as readings drift. This directly targets the three challenges the authors identify: variability of readings, the domain gap in comprehension, and the dynamic nature of sensor data. The system also reports an 11x latency reduction from reusing latent states of cached neighbor readings, and a 503-participant study rates Vivar above several commercial and open generative baselines on coherence, faithfulness, and sensitivity.

What carries the argument

The load-bearing object is the barycentric interpolation map from the sensor-reading space into the CLIP text-embedding space, defined by anchor embeddings and Delaunay tessellation (Equation 1: $E_P = \alpha E_2 + \beta E_3 + \gamma E_4$ for a reading inside triangle $\Delta A_2 A_3 A_4$). The anchors are CLIP text embeddings of sentences describing extreme sensor states, and the Delaunay tessellation partitions the multi-dimensional sensor space into simplices so that every reading has a unique set of non-negative weights. The resulting embedding $E_P$ is decoded by Stable Diffusion through ControlNet with a schema-derived manifestation, and then lifted to 3D through a 3D Gaussian Splatting generator that uses a density encoder to apply controlled dropout so interiors are populated. A cache table of generated images and latent states accelerates updates by seeding new generations with the latent of the nearest cached reading and scaling the number of diffusion iterations by the distance in sensor space.

What would settle it

Compare the interpolated CLIP embedding for an intermediate AQI value (say 157) to the direct CLIP embedding of the literal sentence 'urban skyline with buildings under AQI 157'; if the cosine similarity is markedly lower than the similarities at the two endpoint anchors (44 and 314), the assumption of semantic linearity in the embedding space fails, and the generated image would not be the faithful visualization claimed.

Watch

Extended reading notes

Core claim

The paper claims that a pre-trained visual-semantic embedding space such as CLIP can serve as a faithful bridge between numerical sensor data and generated imagery, provided the mapping is done by anchor-based barycentric interpolation rather than by direct textual prompting of the number. Anchor embeddings are computed by encoding sentences that describe extreme sensor states (e.g., −30°C, 40°C), the sensor space is tessellated with Delaunay triangulation into simplices, and a new reading's embedding is the weighted combination of the vertices of the simplex that contains it. Feeding that interpolated embedding to Stable Diffusion produces images that progress smoothly and monotonically with the sensor value, whereas feeding raw number prompts or averaged embeddings produces inconsistent jumps. The same embedding is then turned into volumetric AR content with a customized 3D Gaussian Splatting pipeline that applies density-aware dropout, and a latent-reuse cache reuses nearby generated latents to cut latency. Vivar is validated by user studies in which it outscores Amazon Titan, DALL-E 3, and two Stable Diffusion variants on coherence, faithfulness, and sensitivity, and by interviews with hydrology and education experts.

Load-bearing premise

The key premise is that interpolating between two anchor embeddings in the CLIP image-text space traces a path of real intermediate sensor states, so the image generator's output for an interpolated embedding is the correct visualization of the in-between reading.

Editorial extensions

If this is right

  • If the embedding-space linearity holds, any sensor type can be visualized without retraining a generator: only a few anchor descriptions and a schema are needed, and the same CLIP-to-diffusion pathway produces the scene.
  • The monotonicity and smoothness of the interpolated embeddings mean that as a sensor reading drifts between two anchor values, the visualization drifts continuously, so users can perceive trends rather than jumps.
  • The latent-reuse cache brings end-to-end generation from roughly 58.5s to 5.3s per update, which makes the system responsive enough for real-time data streams such as air quality or indoor climate.
  • Multi-sensor combinations are handled by simplex interpolation, so temperature, humidity, and other readings jointly determine one coherent scene rather than separate overlays.
  • The weighted user-study scoring (coherence $\omega=0.344$, faithfulness $0.328$, sensitivity $0.328$) suggests user priorities can be folded into a single comparable score, and Vivar's 0.79 overall score is the reported best among the five compared models.

Reading between the lines

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

  • The barycentric-linearity assumption could be probed directly by comparing the interpolated embedding of an intermediate reading against the actual CLIP text embedding of a sentence describing that reading; if those embeddings diverge, the generated image would not be faithful even though it is smooth.
  • The method is a candidate audit tool for pre-trained embedding spaces: the authors' own Figure 3 shows CLIP, VisualBERT, and ALIGN all fail to encode numbers monotonically, and the interpolation bypasses that failure, which suggests the same trick could expose or compensate for digit-bias in other multimodal encoders.
  • The 11x latency gain comes from image-level latent reuse, so the approach would likely transfer to other diffusion backbones and even video generation, where nearby frames share latents.
  • A testable extension is to measure whether the system's accuracy gains survive when anchors are chosen adversarially (e.g., semantically similar but physically opposite extremes), which would stress the assumption that anchor choice is the only free parameter.
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

4 major / 6 minor

Summary. The paper presents Vivar, a generative AR system that maps multi-modal sensor readings into a pre-trained CLIP embedding space via barycentric interpolation over hand-selected anchor embeddings, then uses Stable Diffusion, ControlNet, and 3D Gaussian Splatting to produce 2D and volumetric AR content. The system includes an LLM-based schema and manifestation pipeline and a latent-reuse cache that reduces end-to-end generation time from 58.5 s to 5.3 s. The evaluation includes a user study comparing Vivar with DALL-E 3, Amazon Titan, and two Stable Diffusion variants on coherence, faithfulness, and sensitivity, plus smaller studies and expert interviews.

Significance. If the accuracy claim were properly supported, Vivar would be a useful contribution to HCI for sensor data: it demonstrates a working end-to-end pipeline, a promising use of embedding interpolation for cross-modal visualization, a deployment on commodity AR devices, and a large user study. The paper is weaker as evidence for the central assertion that barycentric interpolation 'accurately reflects' sensor value changes; the quantitative monotonicity result is partly by construction, and the objective quality of the latent-reuse cache is not measured. With additional direct semantic-accuracy tests and statistical reporting, the contribution could meet the bar for a top venue.

major comments (4)
  1. [§5.1, Eq. (1), Fig. 7] Figure 7's Kendall-τ evidence for monotonicity is not independent support for the 'accurate' claim. For unit-norm anchor embeddings E_min and E_max, the cosine similarity of E(t) = (1−t)E_min + tE_max to E_min is monotonically decreasing in t, and to E_max is monotonically increasing, by construction; hence near-perfect τ is expected for any convex interpolation, regardless of whether the decoded image content tracks the sensor value. The only direct visual validation is the single AQI sequence, so the manuscript needs an accuracy test that measures semantic correspondence between generated outputs and the actual sensor reading (e.g., CLIP similarity against a reference text of the true value, or a labeling task with ground-truth values).
  2. [§5.3, Fig. 12] The same convex-combination assumption is extended to Delaunay simplices for multiple sensors, where semantic interactions such as 'cold and humid' need not be linearly composable in CLIP space. Figure 12 compares Vivar with IDW on only one qualitative example, and the 18-participant continuous-frames study reports unanimous preference without any quantitative measure of correctness. Please add a multi-sensor accuracy evaluation: for interior points of a simplex, compare interpolated embeddings to direct CLIP/text embeddings of the corresponding descriptive phrase, and/or ask participants to match generated images to the correct sensor combination among distractors.
  3. [§6.3, Table 1] The claim of '11× latency reduction without compromising quality' is not directly supported: Table 1 reports only timing and iteration reductions, with no evaluation of image or 3DGS output quality under latent reuse versus full generation. Since the efficiency contribution is load-bearing, add a quality comparison (e.g., perceptual metrics, CLIP-score or embedding distance, or a forced-choice user study between cached and uncached outputs) before asserting no quality compromise.
  4. [§7.3, Table 2] Table 2 lacks inferential statistics. The improvements over the second-best model are small (0.02 in faithfulness, 0.05 in sensitivity), and no confidence intervals, significance tests, or raw rating distributions are reported; baseline normalization alone does not establish that these differences are not participant noise. The participant counts are also inconsistent across the paper (abstract: 503; introduction: over 450; §7.3: 565 recruited/485 valid; §7.4: 37; continuous-frames study: 18), which should be reconciled and accompanied by a clear accounting of overlaps.
minor comments (6)
  1. [§5.2] The anchor descriptions are internally inconsistent: A1 is called 'low temperature and low humidity' but given (−30°C, 100%), while A2 is called 'low temperature and high humidity' but given (−30°C, 0%); swap the humidity values.
  2. [§5.1] The subsection titled 'Showcase: Humidity Visualization' actually presents an AQI example ('Urban skyline with buildings under x AQI'); retitle or correct the text.
  3. [Figure 7] The caption reads 'Embedding blinding' (should be 'blending') and the body text refers to 'the top two plots in Figure 7,' but the figure as printed appears to show a single strip; please fix the caption and figure layout.
  4. [Eq. (2)] The weight formula is under-specified: N_Factor, p_Factor,k, and the denominator 'All Factors' are not defined precisely enough to reproduce the reported weights ω_Coherence = 0.344, ω_Faithfulness = 0.328, and ω_Sensitivity = 0.328.
  5. [§7.3] The 'response style adjustment through baseline normalization' is described in one sentence; specify the exact transformation and whether scores were bounded after subtracting the per-participant minimum.
  6. [Figure 3] The motivation experiment would be easier to interpret with a quantitative agreement measure (e.g., Spearman correlation or the slope of the cosine-versus-difference relationship) in addition to the plotted curves.

Circularity Check

1 steps flagged · score 6.0 of 10

Fig. 7's near-perfect Kendall-τ is a mathematical artifact of convex interpolation, so the headline accuracy claim rests on one qualitative AQI example plus subjective user ratings; the monotonicity 'prediction' is built into the construction.

  1. self definitional [Section 5.1, 'Showcase: Humidity Visualization' (Figure 7 and surrounding paragraph)]
    "For interpolation, we use embeddings generated at the minimum and maximum values of x. ... To quantify these differences, we computed the cosine similarity of each embedding relative to the embeddings of the two extreme conditions (AQI of 44 and 314). ... The results indicate that the embedding interpolation method achieves perfect monotonicity, as reflected by Kendall’s Tau values and the smooth progression of the orange curves."

    The interpolated embeddings are convex combinations of exactly the two endpoint anchors, E(t)=(1−t)E_44+tE_314, as formalized for the general case in Eq. (1). For unit-norm CLIP embeddings, cosine similarity to E_44 is monotone decreasing in t and cosine similarity to E_314 is monotone increasing in t, regardless of whether the decoded images semantically match the AQI value. Therefore the reported Kendall-τ=1 is guaranteed by the interpolation construction itself; it cannot serve as independent evidence that the visualization 'accurately reflects value changes.' The only semantic-level validation offered is the single qualitative AQI row in Figure 7, so the quantitative monotonicity result validates the construction against its own definition rather than against external accuracy.

full rationale

Vivar's design intentionally constructs sensor embeddings as convex combinations of anchor CLIP embeddings (Eq. 1, generalized in §5.3). The paper's main objective evidence that this 'accurately reflects value changes' is the AQI experiment in Section 5.1/Figure 7: interpolated embeddings are compared with the two endpoint anchors by cosine similarity, and the interpolation method is reported to achieve perfect Kendall-τ monotonicity. That monotonicity is a mathematical consequence of the construction: for E(t)=(1−t)E_a+tE_b with normalized embeddings, cosine similarity to E_a monotonically decreases and similarity to E_b monotonically increases in t, independent of whether generated image content tracks the sensor reading. Thus the quantitative 'accuracy' result is self-definitional. The remaining support is genuine but partly subjective: the 485-participant study rates Vivar highest on perceived sensitivity, faithfulness, and coherence, and the 37-participant interactive study reports users could distinguish changes. These provide independent grounding for perceived quality. The self-citation [22] (the authors' own MobiCom paper) supports the interpolation idea but is corroborated by external sources [16,43] and is not the sole load-bearing evidence, so I do not count it as a separate circular step. Overall, one central quantitative validation reduces by construction, while the user studies provide independent content; this is partial, not total, circularity.

Assumptions & free parameters 4 free parameters · 5 assumptions · 0 invented entities

The paper does not introduce new physical entities such as particles or forces. Its new modules, such as the density encoder, are unsupported and underspecified, and are therefore counted as free parameters rather than invented entities. The central claims rest on several domain assumptions, especially semantic linearity of the CLIP space, that are not independently verified beyond small examples and subjective user ratings.

free parameters (4)
  • Sensor anchor values = e.g., temperature -30C/40C; humidity 0%/100%; AQI 44/314
    Hand-selected extreme readings define the simplices for barycentric interpolation in Section 5.2. All intermediate embeddings are convex combinations of these anchors, so the choice of anchors determines the entire visualization space.
  • Anchor textual descriptions = not reported (e.g., 'extreme cold', 'extreme hot')
    CLIP text embeddings of the anchor descriptions are the endpoints E_i in Eq. (1). The descriptions are chosen by the authors or LLM without a stated selection criterion.
  • Density encoder dropout rate = not reported
    The 3DGS volumetric filling uses a dropout rate 'determined by the sensor schema and readings through a density encoder' (Section 6.2), but the encoder and mapping are unspecified, making the effect on the central AR output a free design choice.
  • Latent reuse distance-to-iterations schedule = 2-10 diffusion iterations for nearby readings
    The number of iterations is set based on the distance between new and cached readings (Section 6.3). This schedule is ad hoc and not justified by a formal analysis.
assumptions (5)
  • domain assumption CLIP embedding space is a valid shared visual-semantic space for sensor values and visual concepts.
    The whole pipeline maps sensor readings into CLIP text embedding space and decodes them with Stable Diffusion and 3DGS (Section 5.1).
  • domain assumption Linear (barycentric) interpolation between anchor embeddings preserves semantic meaning of intermediate sensor states.
    Eq. (1) computes E_P as a convex combination of anchor embeddings and assumes the result corresponds to the actual reading P (Section 5.2).
  • domain assumption Generative models (Stable Diffusion, ControlNet, DreamGaussian) produce faithful visualizations from CLIP embeddings and sensor schema.
    The visual production stage relies on these pretrained models to reflect sensor values accurately (Sections 6.1 and 6.2).
  • domain assumption LLM-generated sensor schemas and manifestations are valid and consistent.
    The system uses Llama 3.1 to create metadata descriptions and manifestations without task-specific training (Section 6.2).
  • standard math Delaunay tessellation and barycentric coordinates provide a unique, continuous interpolation in n-dimensional sensor space.
    This is standard computational geometry, used in Sections 5.2 and 5.3 to define triangles and tetrahedra.

how reviews work

0 comments
Cite this review

Pith. "Pith review of Visualizing the Invisible: A Generative AR System for Intuitive Multi-Modal Sensor Data Presentation." pith.science (2026). https://pith.science/paper/DBDXXTZE

@misc{pith2026241213509,
  author       = {Pith},
  title        = {Pith review of: Visualizing the Invisible: A Generative AR System for Intuitive Multi-Modal Sensor Data Presentation},
  year         = {2026},
  howpublished = {\url{https://pith.science/paper/DBDXXTZE}},
  note         = {Machine review of arXiv:2412.13509}
}
read the original abstract

Understanding sensor data can be difficult for non-experts because of the complexity and different semantic meanings of sensor modalities. This leads to a need for intuitive and effective methods to present sensor information. However, creating intuitive sensor data visualizations presents three key challenges: the variability of sensor readings, gaps in domain comprehension, and the dynamic nature of sensor data. To address these issues, we propose Vivar, a novel system that integrates multi-modal sensor data and presents 3D volumetric content for AR visualization. In particular, we introduce a cross-modal embedding approach that maps sensor data into a pre-trained visual embedding space through barycentric interpolation. This approach accurately reflects value changes in multi-modal sensor information, ensuring that sensor variations are properly shown in visualization outcomes. Vivar also incorporates sensor-aware AR scene generation using foundation models and 3D Gaussian Splatting (3DGS) without requiring domain expertise. In addition, Vivar leverages latent reuse and caching strategies to accelerate 2D and AR content generation, demonstrating 11x latency reduction without compromising quality. A user study involving over 503 participants, including domain experts, demonstrates Vivar's effectiveness in accuracy, consistency, and real-world applicability, paving the way for more intuitive sensor data visualization.

Figures

Figures reproduced from arXiv: 2412.13509 by the authors.

Figure 1
Figure 1. Left: without Vivar, sensor data are abstract [PITH_FULL_IMAGE:figures/full_fig_p001_1.png] view at source ↗
Figure 2
Figure 2. Survey About Sensor Data Interpretation Context: The temperature is x Celsius 0 20 40 60 80 100 Numerical Difference 0.600 0.700 0.800 0.900 1.000 Cosine Similarity 0 20 40 60 80 100 Numerical Difference 0.990 0.993 0.995 0.998 1.000 Cosine Similarity 0 20 40 60 80 100 Numerical Difference 0.992 0.994 0.996 0.998 1.000 Cosine Similarity (a) CLIP (b) VisualBERT (c) ALIGN [PITH_FULL_IMAGE:figures/full_fig_p003_2.png] view at source ↗
Figure 3
Figure 3. Relationship between cosine similarity and [PITH_FULL_IMAGE:figures/full_fig_p003_3.png] view at source ↗
Figures from the paper (7 more)
Figure 4
Figure 4. Figure 4: Overview of Vivar Workflow. Temporal Dynamics: Sensor data frequently change over time, necessitating visualizations that adapt to real-time changes. To address these challenges, we present Vivar, a system transforming abstract sensor data into intuitive and im￾mersive…
Figure 5
Figure 5. Figure 5: Anchors for embedding sensor readings. presentations. This principle also extends to combinations of readings, ensuring coherence and intuitive interpretation. 5.1 CLIP as the Sensor Visual Space The embedding space must satisfy the following require￾ments: First, it s…
Figure 7
Figure 7. Figure 7: Embedding blinding enhances the monotonic [PITH_FULL_IMAGE:figures/full_fig_p006_7.png]
Figure 9
Figure 9. Figure 9: Sensor embedding to AR visualization. throughout the space. Additionally, the method scales efficiently with the number of sensors and anchors, enabling reliable interpolation across the different sensor modalities. 6 SENSOR-AWARE VISUAL PRODUCTION Section 5 demonstrat…
Figure 10
Figure 10. Figure 10: Comparison of our approach: the Stable Diffusion baseline (top), without manifestation (middle), and Vivar 2D generation with ‘A Calm Room’ manifestation (bottom). and (2) Context: Metadata about environment, purpose, subject, and units. Manifestations. Manifestations…
Figure 13
Figure 13. Figure 13: Model performance comparison across contexts (left) and user survey consid￾erations (right). Model Coherence Faithfulness Sensitivity Overall Score ATitan 0.65 0.68 0.64 0.66 DallE3 0.62 0.70 0.78 0.70 SDE 0.53 0.60 0.60 0.58 SDP 0.59 0.48 0.53 0.53 Vivar (Ours) 0.83 …
Figure 14
Figure 14. Figure 14: Vivar user interface and AR setup. Multi-modality Handling [PITH_FULL_IMAGE:figures/full_fig_p011_14.png]

Discussion (0). Continue with ORCID to comment.

Reference graph

Works this paper leans on

73 extracted references · 37 canonical work pages

  1. [2]

    Jongbaek An, Seungkeun Yeom, Taehoon Hong, Kwangbok Jeong, Jaewook Lee, Scott Eardley, and Junki Choi. 2024. Analysis of the impact of energy consumption data visualization using augmented reality on energy consumption and indoor environ- ment quality. Building and Environment 250 (2024), 111177. https://doi.org/10.1016/j.buildenv.2024.111177

    arXiv 2024

  2. [3]

    Refik Anadol. n.d.. Refik Anadol Studio. https://refikanadol.com/ Accessed: 2024-12-10

    work page 2024

  3. [4]

    Edward Angel. 2005. Interactive computer graphics: a top-down approach using OpenGL. Addison-Wesley Longman Publishing Co., Inc

    work page 2005

  4. [5]

    Siamak Aram, Amedeo Troiano, and Eros Pasero. 2012. Environ- ment sensing using smartphone. In 2012 IEEE Sensors applications symposium proceedings. IEEE, 1–4

    work page 2012

  5. [6]

    Murat Bakirci. 2024. Evaluating the impact of unmanned aerial vehicles (UAVs) on air quality management in smart cities: A comprehensive analysis of transportation-related pol- lution. Computers and Electrical Engineering 119 (2024), 109556. https://doi.org/10.1016/j.compeleceng.2024.109556

    arXiv 2024

  6. [7]

    James Betker, Gabriel Goh, Li Jing, Tim Brooks, Jianfeng Wang, Linjie Li, Long Ouyang, Juntang Zhuang, Joyce Lee, Yufei Guo, et al. 2023. Improving image generation with better captions.Computer Science. https://cdn. openai. com/papers/dall-e-3. pdf 2, 3 (2023), 8

    work page 2023

  7. [8]

    Arpit Bhatia, Henning Pohl, Teresa Hirzle, Hasti Seifi, and Kasper Hornbæk. 2024. Using the Visual Language of Comics to Alter Sensa- tions in Augmented Reality. InProceedings of the CHI Conference on Human Factors in Computing Systems (Honolulu, HI, USA)(CHI ’24). Association for Computing Machinery, New York, NY, USA, Article 603, 17 pages. https://doi....

    arXiv 2024

  8. [9]

    2010–2024

    Ricardo Cabello. 2010–2024. Three.js: A JavaScript 3D Library. https://threejs.org/ Accessed: 2024-09-01

    work page 2010

Show all 73 references
  1. [11]

    Zilong Chen, Feng Wang, Yikai Wang, and Huaping Liu. 2024. Text-to-3d using gaussian splatting. InProceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition. 21401– 21412

    2024

  2. [12]

    Hoi Yau Rosalyn Chun, Yunhan Gao, Rahma Khairunisa Nursalamah, Conor Michael O Keeffe, and Haeji Shin. 2023. CO2LLAB: Creating an Eco-Conscious Community through Habit Tracking and Aug- mented Reality Visualisation. In Extended Abstracts of the 2023 CHI Conference on Human Fac...

    2023

  3. [13]

    Sapers, Noah Deutsch, Malika Khurana, John S

    Adam Coscia, Haley M. Sapers, Noah Deutsch, Malika Khurana, John S. Magyar, Sergio A. Parra, Daniel R. Utter, Rebecca L. Wipfler, David W. Caress, Eric J. Martin, Jennifer B. Paduan, Maggie Hendrie, Santiago Lombeyda, Hillary Mushkin, Alex Endert, Scott Davidoff, and Victoria ...

    2024

  4. [15]

    Patrick Esser, Sumith Kulal, Andreas Blattmann, Rahim Entezari, Jonas Müller, Harry Saini, Yam Levi, Dominik Lorenz, Axel Sauer, Frederic Boesel, et al . 2024. Scaling rectified flow transformers for high-resolution image synthesis. In Forty-first International Conference on M...

    2024

  5. [16]

    Hugging Face. 2023. Images Interpolation with Stable Diffusion. https://github.com/huggingface/cookbook/blob/main/notebooks/en/ stable_diffusion_interpolation.ipynb Accessed: 2023-11-14

    2023

  6. [17]

    Rohit Girdhar, Alaaeldin El-Nouby, Zhuang Liu, Mannat Singh, Kalyan Vasudev Alwala, Armand Joulin, and Ishan Misra. 2023. Im- agebind: One embedding space to bind them all. InProceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition. 15180–15190

    2023

  7. [18]

    Peizhen Gong, Ying Lu, Ruggiero Lovreglio, Xiaofeng Lv, and Zexun Chi. 2024. Applications and effectiveness of augmented reality in safety training: A systematic literature review and meta-analysis. Safety Science 178 (2024), 106624. https://doi.org/10.1016/j.ssci.2024.106624

    2024

  8. [19]

    Branko Grünbaum, Victor Klee, Micha A Perles, and Geoffrey Colin Shephard. 1967. Convex polytopes. Vol. 16. Springer

    1967

  9. [20]

    Jiaming Han, Kaixiong Gong, Yiyuan Zhang, Jiaqi Wang, Kaipeng Zhang, Dahua Lin, Yu Qiao, Peng Gao, and Xiangyu Yue. 2024. Onellm: One framework to align all modalities with language. In Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition. 26584–26595

    2024

  10. [21]

    Carolin Helbig, Maximilian Ueberham, Anna Maria Becker, Heike Marquart, and Uwe Schlink. 2021. Wearable sensors for human environmental exposure in urban settings.Current Pollution Reports 7, 3 (2021), 417–433

    2021

  11. [22]

    Kaiyuan Hou, Yunqi Guo, Heming Fu, Hongkai Chen, Zhenyu Yan, Guoliang Xing, and Xiaofan Jiang. 2024. Improving On-Device LLMs’ Sensory Understanding with Embedding Interpolations. In Proceedings of the 30th Annual International Conference on Mobile Computing and Networking (Wa...

    2024

  12. [23]

    Youyang Hu, Cyprien Raymi Fol, Chiaochi Chou, Verena C Griess, and Yasuaki Kakehi. 2024. Immersive Flora: Re-Engaging with the Forest through the Visualisation of Plant-Environment Interactions in Virtual Reality. InExtended Abstracts of the 2024 CHI Conference on Human Factor...

    2024

  13. [24]

    Le, Yunhsuan Sung, Zhen Li, and Tom Duerig

    Chao Jia, Yinfei Yang, Ye Xia, Yi-Ting Chen, Zarana Parekh, Hieu Pham, Quoc V. Le, Yunhsuan Sung, Zhen Li, and Tom Duerig

  14. [25]

    Yuhao Kang, Fan Zhang, Song Gao, Hui Lin, and Yu Liu. 2020. A review of urban physical environment sensing using street view imagery in public health studies. Annals of GIS 26, 3 (2020), 261–275

    2020

  15. [26]

    Bernhard Kerbl, Georgios Kopanas, Thomas Leimkühler, and George Drettakis. 2023. 3D Gaussian Splatting for Real-Time Radiance Field Rendering. ACM Transactions on Graphics 42, 4 (July 2023). https://repo-sam.inria.fr/fungraph/3d-gaussian-splatting/

    2023

  16. [27]

    Yoonji Kim, Junyi Zhu, Mihir Trivedi, Dishita Turakhia, Ngai Hang Wu, Donghyeon Ko, Michael Wessely, and Stefanie Mueller. 2022. SensorViz: Visualizing Sensor Data Across Different Stages of Prototyping Interactive Objects. In Proceedings of the 2022 ACM Designing Interactive ...

    2022

  17. [28]

    JeongGil Ko, Chenyang Lu, Mani B Srivastava, John A Stankovic, Andreas Terzis, and Matt Welsh. 2010. Wireless sensor networks for healthcare. Proc. IEEE 98, 11 (2010), 1947–1960

    2010

  18. [29]

    Albrecht Kurze, Andreas Bischof, Sören Totzauer, Michael Storz, Maximilian Eibl, Margot Brereton, and Arne Berger. 2020. Guess the Data: Data Work to Understand How People Make Sense of and Use Simple Sensor Data from Homes. InProceedings of the 2020 CHI Conference on Human Fa...

    2020

  19. [30]

    Kevin Kwok. 2023. 3D Gaussian Splatting Implementation. https://github.com/antimatter15/splat. Accessed: 2024-09-02

    2023

  20. [31]

    Der-Tsai Lee and Bruce J Schachter. 1980. Two algorithms for constructing a Delaunay triangulation. International Journal of Computer & Information Sciences 9, 3 (1980), 219–242

    1980

  21. [32]

    Changdi Li, Lei Yu, and Shumin Fei. 2020. Large-Scale, Real-Time 3D Scene Reconstruction Using Visual and IMU Sensors. IEEE Sensors Journal 20, 10 (2020), 5597–5605. https://doi.org/10.1109/JSEN.2020.2971521

    2020

  22. [33]

    Liunian Harold Li, Mark Yatskar, Da Yin, Cho-Jui Hsieh, and Kai-Wei Chang. 2019. VisualBERT: A Simple and Performant Baseline for Vision and Language. arXiv:1908.03557 [cs.CV] https://arxiv.org/abs/1908.03557

    2019 arXiv

  23. [34]

    Chen-Hsuan Lin, Jun Gao, Luming Tang, Towaki Takikawa, Xiaohui Zeng, Xun Huang, Karsten Kreis, Sanja Fidler, Ming-Yu Liu, and Tsung-Yi Lin. 2023. Magic3d: High-resolution text-to-3d content creation. In Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Reco...

    2023

  24. [35]

    Pinyao Liu, Alexandra Kitson, Claudia Picard-Deland, Michelle Carr, Sijia Liu, Ray Lc, and Chen Zhu-Tian. 2024. Virtual Dream Reliving: Exploring Generative AI in Immersive Environment for Dream Re-experiencing. In Extended Abstracts of the 2024 CHI Conference on Human Factors...

    2024

  25. [36]

    Ruoshi Liu, Rundi Wu, Basile Van Hoorick, Pavel Tokmakov, Sergey Zakharov, and Carl Vondrick. 2023. Zero-1-to-3: Zero-shot One Image to 3D Object. arXiv:2303.11328 [cs.CV]

    2023 arXiv

  26. [37]

    Vivian Liu. 2023. Beyond Text-to-Image: Multimodal Prompts to Explore Generative AI. In Extended Abstracts of the 2023 CHI Conference on Human Factors in Computing Systems (Hamburg, Ger- many) (CHI EA ’23). Association for Computing Machinery, New York, NY, USA, Article 482, 6...

    2023

  27. [38]

    Javagar Mahendran, Francesca Schenkel, Ilona Rolfes, and Christian Schulz. 2024. Radar-based Investigation of Elec- tromagnetic Waves Under Different Temperature and Hu- midity Conditions. In 2024 International Conference on Electromagnetics in Advanced Applications (ICEAA). 6...

    2024

  28. [39]

    Malkawi and R

    A. Malkawi and R. Srinivasan. 2005. A new paradigm for Human-Building Interaction: the use of CFD and Aug- mented Reality. Automation in Construction 14 (2005), 71–84. https://doi.org/10.1016/J.AUTCON.2004.08.001

    2005 doi

  29. [40]

    Fintan McGee, Roderick McCall, and Joan Baixauli. 2024. Comparison of Spatial Visualization Techniques for Radiation in Augmented Reality. In Proceedings of the CHI Conference on Human Factors in Computing Systems (Honolulu, HI, USA) (CHI ’24). Association for Computing Machin...

    2024

  30. [41]

    Carolina Nobre, Kehang Zhu, Eric Mörth, Hanspeter Pfister, and Johanna Beyer. 2024. Reading Between the Pixels: Investigating the Barriers to Visualization Literacy. In Proceedings of the CHI Conference on Human Factors in Computing Systems (Honolulu, HI, USA) (CHI ’24). Assoc...

    2024

  31. [42]

    Rita Paradiso. 2003. Wearable health care system for vital signs monitoring. In4th International IEEE EMBS Special TopicConference on Information Technology Applications in Biomedicine, 2003. IEEE, 283–286

    2003

  32. [43]

    Dustin Podell, Zion English, Kyle Lacey, Andreas Blattmann, Tim Dockhorn, Jonas Müller, Joe Penna, and Robin Rombach. 2023. Sdxl: Improving latent diffusion models for high-resolution image synthesis. arXiv preprint arXiv:2307.01952 (2023)

    2023 arXiv

  33. [44]

    Ben Poole, Ajay Jain, Jonathan T Barron, and Ben Mildenhall

  34. [45]

    Ghulam Jilani Quadri, Arran Zeyu Wang, Zhehao Wang, Jennifer Adorno, Paul Rosen, and Danielle Albers Szafir. 2024. Do You See What I See? A Qualitative Study Eliciting High-Level Visualization Com- prehension. In Proceedings of the CHI Conference on Human Factors in Computing ...

    2024

  35. [46]

    Alec Radford, Jong Wook Kim, Chris Hallacy, and et al. 2021. Learning transferable visual models from natural language supervision.arXiv preprint arXiv:2103.00020 (2021)

    2021 arXiv

  36. [47]

    Aditya Ramesh, Prafulla Dhariwal, Alex Nichol, Casey Chu, and Mark Chen. 2022. Hierarchical text-conditional image generation with clip latents. arXiv preprint arXiv:2204.06125 1, 2 (2022), 3

    2022 arXiv

  37. [48]

    Patrick Reipschläger, Frederik Brudy, Raimund Dachselt, Justin Matejka, George Fitzmaurice, and Fraser Anderson. 2022. AvatAR: An Immersive Analysis Environment for Human Motion Data Combining Interactive 3D Avatars and Trajectories. InProceedings of the 2022 CHI Conference on...

    2022

  38. [49]

    Patrick Reipschläger, Tamara Flemisch, and Raimund Dachselt

  39. [50]

    Ivan Rinaldi, Nicola Fanelli, Giovanna Castellano, and Gen- naro Vessio. 2024. Art2Mus: Bridging Visual Arts and Music through Cross-Modal Generation. arXiv:2410.04906 [cs.MM] https://arxiv.org/abs/2410.04906

    2024 arXiv

  40. [51]

    Robin Rombach, Andreas Blattmann, Dominik Lorenz, Patrick Esser, and Björn Ommer. 2022. High-resolution image synthesis with latent diffusion models. In Proceedings of the IEEE/CVF conference on Visualizing the Invisible: A Generative AR System for Intuitive Multi-Modal Sensor...

    2022

  41. [52]

    Mose Sakashita, Balasaravanan Thoravi Kumaravel, Nicolai Mar- quardt, and Andrew David Wilson. 2024. SharedNeRF: Leveraging Photorealistic and View-dependent Rendering for Real-time and Remote Collaboration. In Proceedings of the CHI Conference on Human Factors in Computing Sy...

    2024

  42. [53]

    Harpreet Sareen, Jiefu Zheng, and Pattie Maes. 2019. Cyborg Botany: Augmented Plants as Sensors, Displays and Actuators. In Extended Abstracts of the 2019 CHI Conference on Human Factors in Computing Systems (Glasgow, Scotland Uk) (CHI EA ’19). Association for Computing Machin...

    2019

  43. [54]

    Sarah Schömbs, Jiahe Pan, Yan Zhang, Jorge Goncalves, and Wafa Johal

  44. [55]

    Amazon Web Services. n.d.. Amazon Titan Models. https://docs.aws. amazon.com/bedrock/latest/userguide/titan-models.html Accessed: 2024-12-10

    2024

  45. [56]

    Dennis Sprute, Florian Hufen, Tim Westerhold, and Holger Flatt. 2023. 3D-LiDAR-based Pedestrian Detection for Demand-Oriented Traffic Light Control. International Conference on Industrial Informatics (2023). https://doi.org/10.1109/indin51400.2023.10218109

    2023

  46. [57]

    Jiaxiang Tang, Jiawei Ren, Hang Zhou, Ziwei Liu, and Gang Zeng

  47. [58]

    Aechelon Technology. 2023. U.S. Air Force Reduces Training Footprint with Mixed Reality Multi-Spectral Training Solutions. https://aechelon.com/case-studies/

    2023

  48. [59]

    Gleb Tkachev, Rene Cutura, Michael Sedlmair, Steffen Frey, and Thomas Ertl. 2022. Metaphorical Visualization: Mapping Data to Familiar Concepts. InExtended Abstracts of the 2022 CHI Conference on Human Factors in Computing Systems (New Orleans, LA, USA) (CHI EA ’22). Associati...

    2022

  49. [60]

    David F Watson. 1981. Computing the n-dimensional Delaunay tessellation with application to Voronoi polytopes. The computer journal 24, 2 (1981), 167–172

    1981

  50. [61]

    Wikipedia contributors. 2024. WebXR — Wikipedia, The Free Encyclopedia. https://en.wikipedia.org/wiki/WebXR [Online; accessed 13-Sep-2024]

    2024

  51. [62]

    Rea, and Scott Bateman

    Connor Wilson, Daniel J. Rea, and Scott Bateman. 2023. Temperature Illusions in Mixed Reality using Color and Dynamic Graphics. In Graphics Interface 2023. https://openreview.net/forum?id= U5zU7yDUhKx

    2023

  52. [63]

    Tomoko Yonezawa, Brian Clarkson, Michiaki Yasumura, and Kenji Mase

  53. [64]

    Biao Zhang, Jiapeng Tang, Matthias Niessner, and Peter Wonka

  54. [65]

    Lvmin Zhang, Anyi Rao, and Maneesh Agrawala. 2023. Adding Conditional Control to Text-to-Image Diffusion Models

    2023

  55. [66]

    Yiyuan Zhang, Kaixiong Gong, Kaipeng Zhang, Hongsheng Li, Yu Qiao, Wanli Ouyang, and Xiangyu Yue. 2023. Meta-transformer: A unified framework for multimodal learning. arXiv preprint arXiv:2307.10802 (2023)

    2023 arXiv

  56. [67]

    Minghui Zhao, Stephen Xia, Jingping Nie, Kaiyuan Hou, Avik Dhupar, and Xiaofan Jiang. 2023. LegoSENSE: An Open and Modular Sensing Platform for Rapidly-Deployable IoT Applications. International Conference on Internet-of-Things Design and Implementation (2023). https://doi.org...

    2023

  57. [68]

    Xiaoyu Zhou, Jingqi Wang, Yongtao Wang, Yufei Wei, Nan Dong, and Ming-Hsuan Yang. 2025. AutoOcc: Automatic Open-Ended Semantic Occupancy Annotation via Vision-Language Guided Gaussian Splatting. arXiv:2502.04981 [cs.CV] https://arxiv.org/abs/2502.04981

    2025 arXiv

  58. [69]

    Yue Zhu, Zhiyuan Zhou, Jinlin Miao, Haipeng Mi, and Yijie Guo. 2024. TangibleNegotiation: Probing Design Opportunities for Integration of Generative AI and Swarm Robotics for Imagination Cultivation in Child Art Education. InCompanion of the 2024 on ACM International Joint Con...

    2024

  59. [71]

    arXiv:2301.11445 [cs.CV] https://arxiv.org/abs/2301.11445

    3DShape2VecSet: A 3D Shape Representation for Neural Fields and Generative Diffusion Models. arXiv:2301.11445 [cs.CV] https://arxiv.org/abs/2301.11445

    arXiv

  60. [2001]

    InCHI ’01Extended AbstractsonHumanFactorsinComputingSystems(Seat- tle, Washington)(CHI EA ’01)

    Context-aware sensor-doll as a music expression device. InCHI ’01Extended AbstractsonHumanFactorsinComputingSystems(Seat- tle, Washington)(CHI EA ’01). Association for Computing Machinery, New York, NY, USA, 307–308. https://doi.org/10.1145/634067.634249

  61. [2020]

    IEEE Transactions on Visualization and Computer Graphics 27 (2020), 1182–1192

    Personal Augmented Reality for Information Visual- ization on Large Interactive Displays. IEEE Transactions on Visualization and Computer Graphics 27 (2020), 1182–1192. https://doi.org/10.1109/TVCG.2020.3030460

    2020

  62. [2021]

    Learning With Noisy Text Supervision

    Scaling Up Visual and Vision-Language Representation LastName et al. Learning With Noisy Text Supervision. arXiv:2102.05918 [cs.CV] https://arxiv.org/abs/2102.05918

    arXiv

  63. [2022]

    arXiv preprint arXiv:2209.14988 (2022)

    Dreamfusion: Text-to-3d using 2d diffusion. arXiv preprint arXiv:2209.14988 (2022)

    2022 arXiv

  64. [2023]

    arXiv preprint arXiv:2309.16653 (2023)

    DreamGaussian: Generative Gaussian Splatting for Efficient 3D Content Creation. arXiv preprint arXiv:2309.16653 (2023)

    2023 arXiv

  65. [2024]

    In Extended Abstracts of the 2024 CHI Conference on Human Factors in Computing Systems (CHI EA ’24)

    FaceVis: Exploring a Robot’s Face for Affective Visualisation Design. In Extended Abstracts of the 2024 CHI Conference on Human Factors in Computing Systems (CHI EA ’24). Association for Computing Machinery, New York, NY, USA, Article 161, 10 pages. https://doi.org/10.1145/361...

    2024

Pith tools

Reviewed August 11, 2026 · model on record in the stance chip above.