Pith. sign in

REVIEW 3 major objections 5 minor 66 references

ABO: Abandon Bayer Filter for Adaptive Edge Offloading in Responsive Augmented Reality

T0 review · 3 major / 5 minor · reviewed 2026-08-16 · deepseek-v4-flash

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

Pith's one-line read ABO shows that AR systems can skip demosaicing for offloaded DNN analytics, transmitting compressed RAW tiles instead of RGB frames, and reports 40% higher throughput, 30% lower latency, and up to 15% better DNN accuracy than…

desk verdict Solid RAW offloading system whose headline accuracy gain is really a knowledge-distillation effect; the claimed Bayer-filter benefit is unsupported by the paper's own Table 1. read the letter →

arxiv 2504.20370 v1 pith:KQJTPJ7G submitted 2025-04-29 cs.MM

Yongxuan Han , Shengzhong Liu , Fan Wu , Guihai Chen This is my paper · ORCID
classification cs.MM
keywords augmentedrealityDNNoffloadingRAWimagecompressionneuralcodecBayerfilterdemosaicingadaptivetransmissionedgecomputing
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

ABO claims the conventional color image pipeline works against augmented reality: demosaicing RAW sensor frames into RGB triples the data sent to an edge server for DNN analysis and adds tens of milliseconds of on-device computation, yet does not improve detection accuracy. The paper proposes sending single-channel RAW frames through a lightweight tile-wise neural codec, transmitting only tiles that contained objects in the previous frame, and adapting the codec configuration to current bandwidth, while demosaicing the local rendering stream in parallel. On a prototype AR goggle, the system reports 40% more frame-processing throughput, 30% lower end-to-end latency, and up to 15% higher downstream DNN accuracy than JPEG and three state-of-the-art offloading baselines, and it is the only tested system that reaches the real-time targets of under 40 ms latency and over 25 FPS. If these results generalize, AR devices can get responsive, accurate perception without faster radios or stronger on-board GPUs.

What carries the argument

The load-bearing object is a configurable tile-wise RAW neural codec built as an asymmetric autoencoder. A single convolution layer on the device compresses each selected tile into a feature map; on the server, a deep residual decoder reconstructs the RAW tile. Four configurations are obtained by varying convolution stride (2 or 4) and output channels (4 or 8), and a pluggable decoder head lets one shared decoder serve all four. The RAW frame is disassembled into four CFA color channels before encoding to preserve color information, then reassembled after decoding. Around this codec, a dynamic transmission controller picks tiles by overlap with the previous frame's detection boxes and picks the profiled configuration whose bandwidth cost fits the estimated available bandwidth. Training includes knowledge distillation from the downstream object detector, so the codec learns to keep task-relevant structure rather than only pixel fidelity.

What would settle it

Run ABO against the same baselines on RAW frames from a different sensor (different Bayer pattern, bit depth, or form factor) with a held-out test set from multiple environments; if the mAP gain over JPEG drops below the reported margin, or demosaicing no longer dominates the client-side latency, the central claim fails to generalize.

Watch

Extended reading notes

Core claim

The paper's central discovery is that for machine-centric offloading, the single-channel Bayer RAW frame is the transmitted artifact of choice, not the demosaiced RGB image. ABO shows that a task-aware neural codec can compress RAW tiles to sizes comparable to JPEG while preserving or improving the information the downstream detector needs; knowledge distillation from the detector even sharpens object-relevant edges. Because the RAW stream has one third the channel count and the encoder is a single convolution layer, the client spends little time compressing, and the 25–45 ms demosaicing step can run in parallel with transmission and server inference rather than before them. The resulting pipeline reaches 29 FPS and 37 ms end-to-end latency on an embedded-class client over a 200 Mbps link, and the controller's tile selection plus configuration calibration keeps accuracy high when bandwidth varies. The paper states that among all tested systems, ABO is the only one that met the real-time experience target.

Load-bearing premise

The load-bearing premise is that the self-collected RAW dataset and five consecutive-frame segments, captured with one prototype sensor in campus office scenes, are representative enough that the reported 40% throughput, 30% latency, and 15% accuracy gains transfer to other AR hardware, scenes, lighting, and DNN tasks.

Editorial extensions

If this is right

  • AR offloading can hit the 40 ms and 25 FPS real-time bar on current embedded hardware and a 200 Mbps link, without upgrading the radio or the on-board processor.
  • Machine-centric image codecs trained with downstream task loss dominate human-view codecs like JPEG on the accuracy-bandwidth frontier.
  • Demosaicing can be moved out of the critical path entirely: rendering still gets a full-color frame, but the offloaded analysis never waits for color interpolation.
  • Bandwidth fluctuations can be absorbed by adapting tile selection and codec configuration per frame, keeping task accuracy near the Pareto-optimal frontier.
  • RAW-based offloading degrades less in dim light and fast motion than JPEG-based offloading, which matters when AR users move and rooms are poorly lit.

Reading between the lines

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

  • If the RAW-input advantage holds across sensors, camera ISPs could expose a bypass mode that outputs compressed RAW tiles for machine vision, saving sensor power and silicon area currently spent on demosaicing.
  • The same controller pattern—profiled look-up table, content-aware tile selection, lightweight bandwidth estimate—could extend to other raw sensor streams or to multi-modal AR (audio, depth), though the paper only evaluates image object detection.
  • A quicker test of generality would be to synthesize RAW frames from a public RGB detection benchmark with a different CFA pattern and camera response function; if ABO's 15% accuracy margin does not survive, the gain may be partly dataset-specific.
  • The paper's own limitation notes point to tile-level codec calibration and large-scale RGB pretraining as natural next steps, both of which could reduce bandwidth further and loosen the reliance on large RAW collections.
Share X Bluesky LinkedIn Reddit HN

Signed reviews

No signed human review yet.

Editorial analysis

A structured set of objections, weighed in public.

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

Referee Report

3 major / 5 minor

Summary. The paper proposes ABO, an adaptive offloading framework for AR that compresses pre-demosaic RAW frames with a tile-wise neural codec and a dynamic controller that selects tiles and codec configurations based on estimated bandwidth. The central claim is that skipping demosaicing for offloaded DNN analytics improves end-to-end throughput, latency, and downstream task accuracy relative to standard RGB/JPEG offloading, and the authors support this with a hardware prototype, a self-collected RAW dataset, several baselines, and ablations. The claimed gains are 40% more throughput, 30% less latency, and up to 15% better DNN accuracy than state-of-the-art baselines.

Significance. If validated, the idea of decoupling demosaicing from offloading is a timely and practically relevant contribution to edge-assisted AR, and the throughput/latency savings from 1-channel RAW transmission and pipelining are plausible. The paper also contributes a hardware prototype, a new RAW dataset, and a careful ablation of the controller components. However, the headline accuracy benefit is confounded with task-aware codec training, and the evaluation uses a self-collected dataset without an explicit train/test split, so the causal claim about abandoning the Bayer filter is not yet established. The paper does not release code or data, which limits reproducibility despite its otherwise thorough systems evaluation.

major comments (3)
  1. [Section 2.2, Table 1; Section 3.2.4] The accuracy advantage attributed to RAW offloading is confounded with task-aware distillation. Table 1 shows that the RAW neural codec without knowledge distillation (ABO-noDistill, F1=0.873, mAP=0.878) is less accurate than JPEG (F1=0.883, mAP=0.899), and the advantage of ABO-Distill (F1=0.923, mAP=0.937) appears only after adding the distillation loss of Eq. (2). Since no experiment varies only the input format while keeping the codec and training objective fixed, the title's causal claim that abandoning the Bayer filter improves accuracy is not supported. Please add an RGB-input variant of the same neural codec trained with the same distillation objective, or a demosaiced-RAW variant of ABO, to isolate the effect of the input format from the effect of task-aware codec training.
  2. [Section 5.1.1 and Appendix D] The evaluation uses a self-collected RAW dataset and five consecutive-frame segments, but no explicit train/test split or cross-validation is described. It is unclear whether the five test segments, the frames used to train the codec and the YOLO model, and the 'small profiling data set' used for the LUT in Section 3.3.2 are disjoint. Without this information, the Pareto curves in Figure 10 and the end-to-end results in Figures 11-13 could be optimistically biased by overfitting to one sensor and one scene distribution. Please specify the exact split used for training, profiling, and testing, and, if possible, validate on a second sensor or an independent scene set.
  3. [Abstract and Section 5.3] The claimed 'up to 15%' accuracy improvement in the Abstract is inconsistent with Table 1, where ABO-Distill improves over JPEG by 0.040 F1 and 0.038 mAP (about 4-5% relative). Section 5.3 states that ABO has 'over 15-20% better task accuracy' without defining whether the comparison is absolute or relative, and without identifying the specific operating point on the Pareto curve. Please reconcile these numbers, state the bandwidth at which the 15% figure is achieved, and report variance or confidence intervals for the accuracy measurements.
minor comments (5)
  1. [Section 3.2.1] The text contains the typo 'Appandix C'; it should be 'Appendix C'.
  2. [Section 5.8] The text refers to a 'power moniter'; this should be 'power monitor'.
  3. [Figure 3 caption] The caption contains 'original RA W image' with a stray space; please correct to 'RAW'.
  4. [Appendix G] The limitation statement about limited RAW image data is valuable, but it appears only in the appendix; consider summarizing it in Section 5 so the evaluation's scope is transparent to readers of the main body.
  5. [Section 5.1.1 and Implementation] The paper does not state whether code and data will be released; given the self-collected dataset and prototype, a release statement would substantially aid reproducibility.

Circularity Check

0 steps flagged · score 0.0 of 10

No circularity: the claimed gains are empirical system measurements, not definitional reductions.

full rationale

The derivations in ABO are engineering arguments, not a formal derivation chain with fitted parameters renamed as predictions. The neural codec is trained with MSE and knowledge-distillation losses (Eqs. 1-2), and the reported accuracy is subsequently measured on collected RAW video segments; the task-aware loss is an optimization objective, not a hidden parameter that forces the evaluation metric. The dynamic transmission controller uses an offline-profiled LUT (Section 3.3.1) to select a codec configuration, which is standard profiling rather than a circular prediction: the selection rule is defined by profiled accuracy/bandwidth, but the reported end-to-end throughput/latency/accuracy under replayed bandwidth traces are independent system measurements. No uniqueness theorem, ansatz, or load-bearing self-citation is used; self-citations to the authors' prior offloading work (DeepCOD/AdaMask) are background related work. The abstract's 'up to 15% accuracy' claim may conflate the effect of abandoning demosaicing with the effect of knowledge distillation (Table 1 shows ABO-noDistill below JPEG), but that is a causal-attribution/confounding concern about evaluation validity, not a circularity in which an output is equivalent to its input by construction.

Assumptions & free parameters 3 free parameters · 4 assumptions · 0 invented entities

This is an empirical systems paper with no theoretical derivation, so the ledger consists of design assumptions and hyperparameters rather than fitted model parameters. The main price paid upstream is the strong domain assumption that RAW frames suffice for general DNN analytics, which is only validated on a single task and a self-collected dataset. The codec configurations and key-frame period are hand-chosen degrees of freedom that shape the measured results.

free parameters (3)
  • Key-frame period l
    Chosen by design to bound error accumulation in tile selection; no specific value is reported, and results may depend on this choice.
  • Codec configuration set (kernel stride and output channels) = strides 2/4, channels 4/8
    The four configurations are hand-picked to span an accuracy-size tradeoff; the exact choice affects the Pareto frontier and the controller's options.
  • LUT profiled accuracy and bandwidth values = per-config measured values
    These are measured on a small profiling set and used to make runtime decisions; they are data-dependent and could bias the adaptive controller if the profiling set does not match deployment conditions.
assumptions (4)
  • domain assumption Bayer RAW frames contain sufficient information for the downstream object detection model.
    The whole approach depends on this premise; it is tested in Table 1 for one dataset and one DNN, but not established for other tasks or sensors.
  • domain assumption Demosaicing does not improve downstream DNN accuracy and can be skipped for machine perception.
    This premise is central to the latency and throughput gains; the paper provides empirical support for object detection but not a general proof.
  • domain assumption Network bandwidth is the dominant bottleneck in the tested AR offloading scenarios.
    The adaptive controller assumes that reducing transmitted bytes under limited bandwidth directly improves latency and throughput; this may not hold in compute-bound or extremely low-latency local settings.
  • domain assumption The edge server has abundant compute to run the heavy decoder and DNN.
    ABO offloads computational cost to the server; this asymmetry is a deliberate design choice but limits applicability to serverless or weak-edge settings.

how reviews work

0 comments
Cite this review

Pith. "Pith review of ABO: Abandon Bayer Filter for Adaptive Edge Offloading in Responsive Augmented Reality." pith.science (2026). https://pith.science/paper/KQJTPJ7G

@misc{pith2026250420370,
  author       = {Pith},
  title        = {Pith review of: ABO: Abandon Bayer Filter for Adaptive Edge Offloading in Responsive Augmented Reality},
  year         = {2026},
  howpublished = {\url{https://pith.science/paper/KQJTPJ7G}},
  note         = {Machine review of arXiv:2504.20370}
}
read the original abstract

Bayer-patterned color filter array (CFA) has been the go-to solution for color image sensors. In augmented reality (AR), although color interpolation (i.e., demosaicing) of pre-demosaic RAW images facilitates a user-friendly rendering, it creates no benefits in offloaded DNN analytics but increases the image channels by 3 times inducing higher transmission overheads. The potential optimization in frame preprocessing of DNN offloading is yet to be investigated. To that end, we propose ABO, an adaptive RAW frame offloading framework that parallelizes demosaicing with DNN computation. Its contributions are three-fold: First, we design a configurable tile-wise RAW image neural codec to compress frame sizes while sustaining downstream DNN accuracy under bandwidth constraints. Second, based on content-aware tiles-in-frame selection and runtime bandwidth estimation, a dynamic transmission controller adaptively calibrates codec configurations to maximize the DNN accuracy. Third, we further optimize the system pipelining to achieve lower end-to-end frame processing latency and higher throughput. Through extensive evaluations on a prototype platform, ABO consistently achieves 40% more frame processing throughput and 30% less end-to-end latency while improving the DNN accuracy by up to 15% than SOTA baselines. It also exhibits improved robustness against dim lighting and motion blur situations.

Figures

Figures reproduced from arXiv: 2504.20370 by the authors.

Figure 2
Figure 2. Typical on-sensor image processing pipeline. [PITH_FULL_IMAGE:figures/full_fig_p002_2.png] view at source ↗
Figure 4
Figure 4. Frame processing time comparison between [PITH_FULL_IMAGE:figures/full_fig_p003_4.png] view at source ↗
Figure 6
Figure 6. The tile-wise encoding of ABO with tile-in-frame selection and multiple encoding configurations. [PITH_FULL_IMAGE:figures/full_fig_p004_6.png] view at source ↗
Figures from the paper (11 more)
Figure 7
Figure 7. Figure 7: Color-preserving input preprocessing. 3.3 Dynamic Transmission Controller To cope with network fluctuations, we design a dynamic transmis￾sion controller to continuously calibrate neural codec configura￾tions at runtime. It contains a codec configuration calibrator tha…
Figure 8
Figure 8. Figure 8: 3D-printed AR offloading testing platform. [PITH_FULL_IMAGE:figures/full_fig_p006_8.png]
Figure 10
Figure 10. Figure 10: Accuracy and frame size profiles. 10 20 30 40 50 60 Average Frame Size (KB) 15 20 25 30 35 Throughput (FPS) Better 10 20 30 40 50 60 Average Frame Size (KB) 30 40 50 60 70 Latency (ms) Better ABO JPEG DeepCOD PNC Reducto [PITH_FULL_IMAGE:figures/full_fig_p007_10.png]
Figure 12
Figure 12. Figure 12: Adaptation accuracy under bandwidth dynamics. [PITH_FULL_IMAGE:figures/full_fig_p007_12.png]
Figure 14
Figure 14. Figure 14: Accuracy degradation under low luminosity. [PITH_FULL_IMAGE:figures/full_fig_p008_14.png]
Figure 17
Figure 17. Figure 17: The configurable image neural codec in ABO. [PITH_FULL_IMAGE:figures/full_fig_p011_17.png]
Figure 18
Figure 18. Figure 18: Random Generated Bandwidth Trace 1 [PITH_FULL_IMAGE:figures/full_fig_p011_18.png]
Figure 20
Figure 20. Figure 20: Real-World Recorded Bandwidth Trace 1. 0 5 10 15 20 25 30 Time (s) 0 4 8 12 16 20 Bandwidth (Mbps) [PITH_FULL_IMAGE:figures/full_fig_p011_20.png]
Figure 22
Figure 22. Figure 22: Example of Low-Light Frames Clear Image High-Motion [PITH_FULL_IMAGE:figures/full_fig_p012_22.png]
Figure 23
Figure 23. Figure 23: Example of High-Motion Frames Seg3 Seg5 15 25 20 30 Throughput (FPS) Seg3 Seg5 35 45 55 65 Latency (ms) ABO JPG DeepCOD PNC Reducto [PITH_FULL_IMAGE:figures/full_fig_p012_23.png]
Figure 24
Figure 24. Figure 24: Efficiency under high-motion sample from the trace at the corresponding time. We visualize two of the randomly generated traces in [PITH_FULL_IMAGE:figures/full_fig_p012_24.png]

Discussion (0). Continue with ORCID to comment.

Reference graph

Works this paper leans on

66 extracted references · 54 canonical work pages

  1. [1]

    Google Cardboard

    2014. Google Cardboard. https://arvr.google.com/cardboard

    work page 2014

  2. [2]

    Ubuntu tc8

    2019. Ubuntu tc8. https://manpages.ubuntu.com/manpages/focal/man8/tc.8.html

    work page 2019

  3. [3]

    NVIDIA TensorRT

    2024. NVIDIA TensorRT. https://developer.nvidia.com/tensorrt

    work page 2024

  4. [4]

    Ganesh Ananthanarayanan, Paramvir Bahl, Peter Bodík, Krishna Chintalapudi, Matthai Philipose, Lenin Ravindranath, and Sudipta Sinha. 2017. Real-time video analytics: The killer app for edge computing. computer 50, 10 (2017), 58–67

    work page 2017

  5. [5]

    Kevin Brownlow. 1980. What Was the Right Speed? https://cinemaweb.com/ silentfilm/bookshelf/18_kb_2.htm

    work page 1980

  6. [6]

    Chen Chen, Qifeng Chen, Jia Xu, and Vladlen Koltun. 2018. Learning to see in the dark. In Proceedings of the IEEE conference on computer vision and pattern recognition. 3291–3300

    work page 2018

  7. [7]

    Yihua Cheng, Ziyi Zhang, Hanchen Li, Anton Arapin, Yue Zhang, Qizheng Zhang, Yuhan Liu, Kuntai Du, Xu Zhang, Francis Y Yan, et al. 2024. GRACE: Loss- Resilient Real-Time Video through Neural Codecs. In 21st USENIX Symposium on Networked Systems Design and Implementation (NSDI 24) . 509–531

    work page 2024

  8. [8]

    Zhengxue Cheng, Heming Sun, Masaru Takeuchi, and Jiro Katto. 2020. Learned image compression with discretized gaussian mixture likelihoods and attention modules. In Proceedings of the IEEE/CVF conference on computer vision and pattern recognition. 7939–7948

    2020

Show all 66 references
  1. [9]

    Kiyoyuki Chinzei, Nobuhiko Hata, Ferenc A Jolesz, and Ron Kikinis. 2000. MR compatible surgical assist robot: System integration and preliminary feasibility study. In Medical Image Computing and Computer-Assisted Intervention–MICCAI 2000: Third International Conference, Pittsb...

    2000

  2. [10]

    King-Hong Chung and Yuk-Hee Chan. 2008. A lossless compression scheme for Bayer color filter array images. IEEE Transactions on Image Processing 17, 2 (2008), 134–144

    2008

  3. [11]

    Jia Deng, Wei Dong, Richard Socher, Li-Jia Li, Kai Li, and Li Fei-Fei. 2009. Imagenet: A large-scale hierarchical image database. In 2009 IEEE conference on computer vision and pattern recognition . Ieee, 248–255

    2009

  4. [12]

    Xingbo Dong, Wanyan Xu, Zhihui Miao, Lan Ma, Chao Zhang, Jiewen Yang, Zhe Jin, Andrew Beng Jin Teoh, and Jiajun Shen. 2022. Abandoning the bayer-filter to see in the dark. In Proceedings of the ieee/cvf conference on computer vision and pattern recognition. 17431–17440

    2022

  5. [13]

    Amir Erfan Eshratifar and Massoud Pedram. 2018. Energy and performance efficient computation offloading for deep neural networks in a mobile cloud computing environment. In Proceedings of the 2018 on Great Lakes Symposium on VLSI. 111–116

    2018

  6. [14]

    Mark Everingham, SM Ali Eslami, Luc Van Gool, Christopher KI Williams, John Winn, and Andrew Zisserman. 2015. The pascal visual object classes challenge: A retrospective. International journal of computer vision 111 (2015), 98–136

    2015

  7. [15]

    Yongjie Guan, Xueyu Hou, Nan Wu, Bo Han, and Tao Han. 2023. Metastream: Live volumetric content capture, creation, delivery, and rendering in real time. In Proceedings of the 29th Annual International Conference on Mobile Computing and Networking. 1–15

    2023

  8. [16]

    Kaiming He, Xiangyu Zhang, Shaoqing Ren, and Jian Sun. 2016. Deep residual learning for image recognition. In Proceedings of the IEEE conference on computer vision and pattern recognition . 770–778

    2016

  9. [17]

    Qiang He, Zeqian Dong, Feifei Chen, Shuiguang Deng, Weifa Liang, and Yun Yang. 2022. Pyramid: Enabling hierarchical neural networks with edge computing. In Proceedings of the ACM Web Conference 2022 . 1860–1870

    2022

  10. [18]

    HikRobot. 2024. MV-CB060-10UM-S. https://www.hikrobotics.com/cn/ machinevision/productdetail?id=3634

    2024

  11. [19]

    Geoffrey Hinton. 2015. Distilling the Knowledge in a Neural Network. arXiv preprint arXiv:1503.02531 (2015)

    2015 arXiv

  12. [20]

    Jin Huang, Hui Guan, and Deepak Ganesan. 2023. Re-thinking computation of- fload for efficient inference on IoT devices with duty-cycled radios. InProceedings of the 29th Annual International Conference on Mobile Computing and Networking . 1–15

    2023

  13. [21]

    Monsoon Solutions Inc. 2023. High Voltage Power Monitor. https://www.msoon. com/high-voltage-power-monitor

    2023

  14. [22]

    Nuowen Kan, Yuankun Jiang, Chenglin Li, Wenrui Dai, Junni Zou, and Hongkai Xiong. 2022. Improving generalization for neural adaptive video streaming via meta reinforcement learning. In Proceedings of the 30th ACM International Conference on Multimedia. 3006–3016

    2022

  15. [23]

    Yiping Kang, Johann Hauswald, Cao Gao, Austin Rovinski, Trevor Mudge, Jason Mars, and Lingjia Tang. 2017. Neurosurgeon: Collaborative intelligence between the cloud and mobile edge. ACM SIGARCH Computer Architecture News 45, 1 (2017), 615–629

    2017

  16. [24]

    R. Kimmel. 1999. Demosaicing: image reconstruction from color CCD samples. IEEE Transactions on Image Processing 8, 9 (1999), 1221–1228. doi:10.1109/83. 784434

    1999 doi

  17. [25]

    Kodak. 1999. Kodak Lossless True Color Image Suite . https://r0k.us/graphics/ kodak/

    1999

  18. [26]

    Z Jonny Kong, Qiang Xu, Jiayi Meng, and Y Charlie Hu. 2023. AccuMO: Accuracy- centric multitask offloading in edge-assisted mobile augmented reality. In Pro- ceedings of the 29th Annual International Conference on Mobile Computing and Networking. 1–16

    2023

  19. [27]

    M Lakshmi, J Senthilkumar, and Y Suresh. 2016. Visually lossless compression for Bayer color filter array using optimized vector quantization. Applied Soft Computing 46 (2016), 1030–1042

    2016

  20. [28]

    Yuanqi Li, Arthi Padmanabhan, Pengzhan Zhao, Yufei Wang, Guoqing Harry Xu, and Ravi Netravali. 2020. Reducto: On-camera filtering for resource-efficient real-time video analytics. In Proceedings of the Annual conference of the ACM Special Interest Group on Data Communication o...

    2020

  21. [29]

    Tsung-Yi Lin, Michael Maire, Serge Belongie, James Hays, Pietro Perona, Deva Ramanan, Piotr Dollár, and C Lawrence Zitnick. 2014. Microsoft coco: Common objects in context. In Computer Vision–ECCV 2014: 13th European Conference, Zurich, Switzerland, September 6-12, 2014, Proce...

    2014

  22. [30]

    Luyang Liu, Hongyu Li, and Marco Gruteser. 2019. Edge assisted real-time object detection for mobile augmented reality. In The 25th annual international conference on mobile computing and networking . 1–16

    2019

  23. [31]

    Shengzhong Liu, Xinzhe Fu, Maggie Wigness, Philip David, Shuochao Yao, Lui Sha, and Tarek Abdelzaher. 2022. Self-cueing real-time attention scheduling in criticality-aware visual machine perception. In 2022 IEEE 28th Real-Time and Embedded Technology and Applications Symposium...

    2022

  24. [32]

    Shengzhong Liu, Tianshi Wang, Jinyang Li, Dachun Sun, Mani Srivastava, and Tarek Abdelzaher. 2022. Adamask: Enabling machine-centric video streaming with adaptive frame masking for dnn inference offloading. In Proceedings of the 30th ACM international conference on multimedia ...

    2022

  25. [33]

    Xingyu Liu, Pengfei Ren, Yuchen Chen, Cong Liu, Jing Wang, Haifeng Sun, Qi Qi, and Jingyu Wang. 2023. SA-Fusion: Multimodal Fusion Approach for Web- based Human-Computer Interaction in the Wild. In Proceedings of the ACM Web Conference 2023. 3883–3891

    2023

  26. [34]

    Zihao Liu, Tao Liu, Wujie Wen, Lei Jiang, Jie Xu, Yanzhi Wang, and Gang Quan

  27. [35]

    Guoyu Lu. 2023. Object Detection Based on Raw Bayer Images. In 2023 IEEE/RSJ International Conference on Intelligent Robots and Systems (IROS) . IEEE, 9582– 9589

    2023

  28. [36]

    Yuzhu Lu and Shana Smith. 2007. Augmented reality e-commerce assistant sys- tem: trying while shopping. InHuman-Computer Interaction. Interaction Platforms and Techniques: 12th International Conference, HCI International 2007, Beijing, China, July 22-27, 2007, Proceedings, Par...

    2007

  29. [37]

    Tianrui Ma, Adith Jagadish Boloor, Xiangxing Yang, Weidong Cao, Patrick Williams, Nan Sun, Ayan Chakrabarti, and Xuan Zhang. 2023. Leca: In-sensor learned compressive acquisition for efficient machine vision on the edge. In Proceedings of the 50th Annual International Symposiu...

    2023

  30. [38]

    Pavel Mach and Zdenek Becvar. 2017. Mobile edge computing: A survey on architecture and computation offloading.IEEE communications surveys & tutorials 19, 3 (2017), 1628–1656

    2017

  31. [39]

    Daniele Menon, Stefano Andriani, and Giancarlo Calvagno. 2006. Demosaicing with directional filtering and a posteriori decision. IEEE Transactions on Image Processing 16, 1 (2006), 132–141

    2006

  32. [40]

    Fabian Mentzer, George D Toderici, Michael Tschannen, and Eirikur Agustsson

  33. [41]

    David Minnen and Saurabh Singh. 2020. Channel-wise autoregressive entropy models for learned image compression. In 2020 IEEE International Conference on Image Processing (ICIP). IEEE, 3339–3343

    2020

  34. [42]

    D Darian Muresan and Thomas W Parks. 2005. Demosaicing using optimal recovery. IEEE Transactions on Image Processing 14, 2 (2005), 267–278

    2005

  35. [43]

    Adam Paszke, Sam Gross, Francisco Massa, and Lerer et al. 2019. Pytorch: An imperative style, high-performance deep learning library. Advances in neural information processing systems 32 (2019)

    2019

  36. [44]

    Wayne Piekarski and Bruce Thomas. 2002. ARQuake: the outdoor augmented reality gaming system. Commun. ACM 45, 1 (2002), 36–38

    2002

  37. [45]

    J Redmon. 2016. You only look once: Unified, real-time object detection. In Proceedings of the IEEE conference on computer vision and pattern recognition

    2016

  38. [46]

    Jie Ren, Ling Gao, Xiaoming Wang, Miao Ma, Guoyong Qiu, Hai Wang, Jie Zheng, and Zheng Wang. 2021. Adaptive computation offloading for mobile augmented reality. Proceedings of the ACM on Interactive, Mobile, Wearable and Ubiquitous Technologies 5, 4 (2021), 1–30

    2021

  39. [47]

    Thomas Richter, Siegfried Fößel, Antonin Descampe, and Gaël Rouvroy. 2021. Bayer CFA pattern compression with JPEG XS. IEEE Transactions on Image Processing 30 (2021), 6557–6569

    2021

  40. [48]

    Allen Rush and Paul Hubel. 2003. X3 sensor characteristics.Journal of The Society of Photographic Science and Technology of Japan 66, 1 (2003), 57–60. WWW ’25, April 28-May 2, 2025, Sydney, NSW, Australia Yongxuan Han et al

    2003

  41. [49]

    Mahesh B Shenai, Marcus Dillavou, Corey Shum, Douglas Ross, Richard S Tubbs, Alan Shih, and Barton L Guthrie. 2011. Virtual interactive presence and aug- mented reality (VIPAR) for remote surgical assistance. Operative Neurosurgery 68 (2011), ons200–ons207

    2011

  42. [50]

    Mingxing Tan and Quoc Le. 2019. Efficientnet: Rethinking model scaling for convolutional neural networks. In International conference on machine learning . PMLR, 6105–6114

    2019

  43. [51]

    George Toderici, Sean M O’Malley, Sung Jin Hwang, Damien Vincent, David Minnen, Shumeet Baluja, Michele Covell, and Rahul Sukthankar. 2015. Vari- able rate image compression with recurrent neural networks. arXiv preprint arXiv:1511.06085 (2015)

    2015 arXiv

  44. [52]

    George Toderici, Damien Vincent, Nick Johnston, Sung Jin Hwang, David Minnen, Joel Shor, and Michele Covell. 2017. Full resolution image compression with recurrent neural networks. In Proceedings of the IEEE conference on Computer Vision and Pattern Recognition . 5306–5314

    2017

  45. [53]

    Gregory K Wallace. 1992. The JPEG still picture compression standard. IEEE transactions on consumer electronics 38, 1 (1992), xviii–xxxiv

    1992

  46. [54]

    Hao Wang, Hao Bao, Liekang Zeng, Ke Luo, and Xu Chen. 2023. Real-Time High- Resolution Pedestrian Detection in Crowded Scenes via Parallel Edge Offloading. In ICC 2023-IEEE International Conference on Communications . IEEE, 2173–2178

    2023

  47. [55]

    Ruiqi Wang, Hanyang Liu, Jiaming Qiu, Moran Xu, Roch Guérin, and Chenyang Lu. 2023. Progressive Neural Compression for Adaptive Image Offloading under Timing Constraints. In 2023 IEEE Real-Time Systems Symposium (RTSS) . IEEE, 118–130

    2023

  48. [56]

    Cliff Wootton. 2005. A practical guide to video and audio compression: From sprockets and rasters to macro blocks . Routledge

    2005

  49. [57]

    Hsin-Kai Wu, Silvia Wen-Yu Lee, Hsin-Yi Chang, and Jyh-Chong Liang. 2013. Current status, opportunities and challenges of augmented reality in education. Computers & education 62 (2013), 41–49

    2013

  50. [58]

    Shuo Xiao, Shengzhi Wang, Zhenzhen Huang, Tianyu Wang, Wei Chen, and Guopeng Zhang. 2021. Task offloading strategy of internet of vehicles based on stackelberg game. In Companion Proceedings of the Web Conference 2021 . 52–56

    2021

  51. [59]

    Xiufeng Xie and Kyu-Han Kim. 2019. Source compression with bounded dnn perception loss for iot edge computer vision. In The 25th Annual International Conference on Mobile Computing and Networking . 1–16

    2019

  52. [60]

    Simin Yang, Ze Gao, Reza Hadi Mogavi, Pan Hui, and Tristan Braud. 2023. Tangi- ble web: An interactive immersion virtual reality creativity system that travels across reality. In Proceedings of the ACM Web Conference 2023 . 3915–3922

    2023

  53. [61]

    Shuochao Yao, Jinyang Li, Dongxin Liu, Tianshi Wang, Shengzhong Liu, Huajie Shao, and Tarek Abdelzaher. 2020. Deep compressive offloading: Speeding up neural network inference by trading edge computation for network latency. In Proceedings of the 18th conference on embedded ne...

    2020

  54. [62]

    Mark Yi-Cheon Yim, Shu-Chuan Chu, and Paul L Sauer. 2017. Is augmented reality technology an effective tool for e-commerce? An interactivity and vividness perspective. Journal of interactive marketing 39, 1 (2017), 89–103

    2017

  55. [63]

    Steve Chi-Yin Yuen, Gallayanee Yaoyuneyong, and Erik Johnson. 2011. Aug- mented reality: An overview and five directions for AR in education. Journal of Educational Technology Development and Exchange (JETDE) 4, 1 (2011), 11

    2011

  56. [64]

    Wuyang Zhang, Zhezhi He, Luyang Liu, Zhenhua Jia, Yunxin Liu, Marco Gruteser, Dipankar Raychaudhuri, and Yanyong Zhang. 2021. Elf: accelerate high-resolution mobile deep vision with content-aware parallel offloading. In Proceedings of the 27th Annual International Conference o...

    2021

  57. [2018]

    In Proceedings of the 55th annual design automation conference

    DeepN-JPEG: A deep neural network favorable JPEG-based image compres- sion framework. In Proceedings of the 55th annual design automation conference . 1–6

  58. [2020]

    High-fidelity generative image compression.Advances in Neural Information Processing Systems 33 (2020), 11913–11924

    2020

Pith tools

Reviewed August 16, 2026 · model on record in the stance chip above.