pith. sign in

arxiv: 1906.10327 · v2 · pith:ANJXBWNWnew · submitted 2019-06-25 · 💻 cs.CV

SkyNet: A Champion Model for DAC-SDC on Low Power Object Detection

Xiaofan Zhang , Cong Hao , Haoming Lu , Jiachen Li , Yuhong Li , Yuchen Fan , Kyle Rupnow , Jinjun Xiong
show 4 more authors
Thomas Huang Honghui Shi Wen-Mei Hwu Deming Chen
This is my paper

Pith reviewed 2026-05-25 17:07 UTC · model grok-4.3

classification 💻 cs.CV
keywords lightweight DNNobject detectionUAVedge computingFPGAGPUconvolutional neural networklow power
0
0 comments X

The pith

SkyNet is a 12-layer DNN with 1.82 MB parameters that won first place in low-power UAV object detection on both GPU and FPGA.

A machine-rendered reading of the paper's core claim, the machinery that carries it, and where it could break.

The paper introduces SkyNet as an extremely lightweight convolutional network built through a bottom-up design process specifically for edge deployment. It reports first-place results in the DAC-SDC contest: 0.731 IoU at 67.33 FPS on a TX2 GPU and 0.716 IoU at 25.05 FPS on an Ultra96 FPGA. A sympathetic reader would care because UAV vision systems must deliver real-time detection while staying within tight power and memory budgets that defeat most standard models. The work demonstrates that a compact network can satisfy both accuracy and throughput targets on embedded hardware without relying on larger, more resource-heavy architectures.

Core claim

SkyNet is an extremely lightweight DNN with 12 convolutional layers and only 1.82 MB of parameters that won the first place award for both the GPU and FPGA tracks of the DAC-SDC low power object detection challenge on UAV images, delivering 0.731 IoU and 67.33 FPS on a TX2 GPU and 0.716 IoU and 25.05 FPS on an Ultra96 FPGA.

What carries the argument

Bottom-up DNN design approach that builds the 12-layer network layer by layer to meet the contest metrics of IoU and FPS under implied power limits.

If this is right

  • Compact networks built this way can support real-time object detection directly on UAV hardware without cloud offload.
  • The same bottom-up construction method can be reused for other edge vision tasks that face similar accuracy-throughput-power trade-offs.
  • Winning entries on the DAC-SDC benchmark indicate that contest-specific optimization can produce models that meet practical drone deployment constraints.

Where Pith is reading between the lines

These are editorial extensions of the paper, not claims the author makes directly.

  • If the bottom-up method generalizes, similar lightweight networks could be derived for additional embedded platforms such as mobile SoCs or custom ASICs.
  • Success on UAV imagery suggests the approach may extend to other constrained vision domains like autonomous ground vehicles or surveillance cameras.
  • The reported parameter count and layer depth provide a concrete baseline for measuring how much further compression remains possible while preserving the achieved IoU-FPS balance.

Load-bearing premise

The model tuned to the contest dataset and two specific hardware platforms will maintain comparable accuracy and speed on other UAV images and edge devices.

What would settle it

Running SkyNet on a fresh collection of UAV images under the same power envelope and observing IoU fall below 0.7 or FPS drop below the reported values on either the TX2 or Ultra96 would falsify the central performance claim.

Figures

Figures reproduced from arXiv: 1906.10327 by Cong Hao, Deming Chen, Haoming Lu, Honghui Shi, Jiachen Li, Jinjun Xiong, Kyle Rupnow, Thomas Huang, Wen-Mei Hwu, Xiaofan Zhang, Yuchen Fan, Yuhong Li.

Figure 1
Figure 1. Figure 1: DAC-SDC Top-3 design comparison of inference [PITH_FULL_IMAGE:figures/full_fig_p001_1.png] view at source ↗
Figure 2
Figure 2. Figure 2: Examples from the training dataset with red color bounding boxes in main categories as rider (a), drone (b), horse [PITH_FULL_IMAGE:figures/full_fig_p002_2.png] view at source ↗
Figure 3
Figure 3. Figure 3: A widely-used top-down DNN design approach for [PITH_FULL_IMAGE:figures/full_fig_p002_3.png] view at source ↗
Figure 4
Figure 4. Figure 4: The bottom-up DNN design flow we adopt from (Hao et al. 2019b) for SkyNet design. Without relying on certain [PITH_FULL_IMAGE:figures/full_fig_p004_4.png] view at source ↗
Figure 7
Figure 7. Figure 7: It clearly shows that 91% of the objects to be de [PITH_FULL_IMAGE:figures/full_fig_p004_7.png] view at source ↗
Figure 5
Figure 5. Figure 5: SkyNet architecture (version C in Table 2) generated by stacking six of the selected Bundle (circled by green dash [PITH_FULL_IMAGE:figures/full_fig_p005_5.png] view at source ↗
Figure 6
Figure 6. Figure 6: Feature map reordering from 1×4×4 to 4×2×2 with shrunken width and height but expanded number of channels. There is no information loss compared to tradi￾tional pooling. In addition, this reorder pattern also ensures larger receptive field [PITH_FULL_IMAGE:figures/full_fig_p005_6.png] view at source ↗
Figure 7
Figure 7. Figure 7: The distribution of bounding box relative size in [PITH_FULL_IMAGE:figures/full_fig_p006_7.png] view at source ↗
read the original abstract

Developing artificial intelligence (AI) at the edge is always challenging, since edge devices have limited computation capability and memory resources but need to meet demanding requirements, such as real-time processing, high throughput performance, and high inference accuracy. To overcome these challenges, we propose SkyNet, an extremely lightweight DNN with 12 convolutional (Conv) layers and only 1.82 megabyte (MB) of parameters following a bottom-up DNN design approach. SkyNet is demonstrated in the 56th IEEE/ACM Design Automation Conference System Design Contest (DAC-SDC), a low power object detection challenge in images captured by unmanned aerial vehicles (UAVs). SkyNet won the first place award for both the GPU and FPGA tracks of the contest: we deliver 0.731 Intersection over Union (IoU) and 67.33 frames per second (FPS) on a TX2 GPU and deliver 0.716 IoU and 25.05 FPS on an Ultra96 FPGA.

Editorial analysis

A structured set of objections, weighed in public.

Desk editor's note, referee report, simulated authors' rebuttal, and a circularity audit. Tearing a paper down is the easy half of reading it; the pith above is the substance, this is the friction.

Referee Report

0 major / 2 minor

Summary. The manuscript presents SkyNet, an extremely lightweight DNN with 12 convolutional layers and 1.82 MB of parameters developed via a bottom-up design approach for low-power object detection on UAV images. It reports that this model won first place in both the GPU and FPGA tracks of the 2019 DAC-SDC contest, delivering 0.731 IoU at 67.33 FPS on a TX2 GPU and 0.716 IoU at 25.05 FPS on an Ultra96 FPGA.

Significance. If the reported contest outcomes are accurate, the work supplies a concrete, hardware-specific benchmark for efficient edge-based object detection under strict power and resource limits. The explicit first-place results on the contest-specified platforms (TX2 and Ultra96) and the small parameter count provide a practical reference point for the community. The paper does not claim or demonstrate generalization beyond the DAC-SDC evaluation dataset and hardware; the central claim remains anchored to the external contest evaluation rather than an internal derivation.

minor comments (2)
  1. The abstract (and likely the methods section) reports the contest-winning IoU and FPS numbers but supplies no information on training procedure, data splits, hyperparameter choices, or error analysis, limiting reproducibility of the model development process.
  2. The description of the bottom-up DNN design approach would benefit from additional concrete steps or pseudocode showing how the 12-layer architecture and 1.82 MB parameter count were derived from the contest metrics.

Simulated Author's Rebuttal

0 responses · 0 unresolved

We thank the referee for the review and the recommendation of minor revision. No specific major comments were provided in the report, and the referee summary accurately restates the manuscript's contributions and contest results. We are pleased that the work is recognized as supplying a practical benchmark for edge-based object detection under the DAC-SDC constraints.

Circularity Check

0 steps flagged

No significant circularity

full rationale

The paper is an empirical report of a contest-winning DNN design (SkyNet) evaluated on the external DAC-SDC benchmark with measured IoU/FPS on specified hardware. No mathematical derivation, fitted parameters renamed as predictions, or self-citation chain is present in the provided text; the central claim rests on contest outcomes rather than internal equations that could reduce to inputs by construction.

Axiom & Free-Parameter Ledger

0 free parameters · 0 axioms · 0 invented entities

The abstract supplies no information on free parameters, axioms, or invented entities; therefore the ledger is empty.

pith-pipeline@v0.9.0 · 5739 in / 1271 out tokens · 38684 ms · 2026-05-25T17:07:23.499437+00:00 · methodology

discussion (0)

Sign in with ORCID, Apple, or X to comment. Anyone can read and Pith papers without signing in.

Reference graph

Works this paper leans on

27 extracted references · 27 canonical work pages · 4 internal anchors

  1. [1]

    write newline

    " write newline "" before.all 'output.state := FUNCTION fin.entry add.period write newline FUNCTION new.block output.state before.all = 'skip after.block 'output.state := if FUNCTION new.sentence output.state after.block = 'skip output.state before.all = 'skip after.sentence 'output.state := if if FUNCTION not #0 #1 if FUNCTION and 'skip pop #0 if FUNCTIO...

    work page

  2. [2]

    Deng, J., and Zhuo, C. a. 2018. DAC-SDC'18 2nd place winner in GPU track. https://github.com/jndeng/DACSDC-DeepZ. Accessed: 2019-06-09

    work page 2018

  3. [3]

    Deng, J.; Dong, W.; Socher, R.; Li, L.-J.; Li, K.; and Fei-Fei, L. 2009. Imagenet: A large-scale hierarchical image database. In 2009 IEEE conference on computer vision and pattern recognition , 248--255. Ieee

    work page 2009

  4. [4]

    Deng, J.; Shen, T.; Yan, X.; Chen, Y.; Zhang, H.; Wang, R.; Zhou, P.; and Zhuo, C. 2019. DAC-SDC'19 3rd place winner in GPU track

    work page 2019

  5. [5]

    DJI. 2018. DAC-SDC dataset. https://github.com/xyzxinyi zhang/2018-DAC-System-Design-Contest. Accessed: 2019-06-08

    work page 2018

  6. [6]

    Franklin, D. 2017. NVIDIA Jetson TX2 delivers twice the intelligence to the edge. NVIDIA Accelerated Computing| Parallel Forall

    work page 2017

  7. [7]

    H.; Zhang, X.; Gao, T.; Xiong, J.; Rupnow, K.; Yu, H.; Hwu, W.-M.; and Chen, D

    Hao, C.; Li, Y.; Huang, S. H.; Zhang, X.; Gao, T.; Xiong, J.; Rupnow, K.; Yu, H.; Hwu, W.-M.; and Chen, D. 2018. DAC-SDC'18 3rd place winner in FPGA track. https://github.com/onioncc/iSmartDNN. Accessed: 2019-06-09

    work page 2018

  8. [8]

    H.; Rupnow, K.; Xiong, J.; Hwu, W.-M.; and Chen, D

    Hao, C.; Zhang, X.; Li, Y.; Chen, Y.; Liu, X.; Huang, S. H.; Rupnow, K.; Xiong, J.; Hwu, W.-M.; and Chen, D. 2019a. DAC-SDC'19 1st place winner in FPGA track

    work page

  9. [9]

    Hao, C.; Zhang, X.; Li, Y.; Huang, S.; Xiong, J.; Rupnow, K.; Hwu, W.-m.; and Chen, D. 2019b. FPGA/DNN co-design: An efficient design methodology for IoT intelligence on the edge. In Proceedings of the 56th Annual Design Automation Conference 2019 , 206. ACM

    work page 2019

  10. [10]

    He, K.; Zhang, X.; Ren, S.; and Sun, J. 2016. Deep residual learning for image recognition. In Proceedings of the IEEE conference on computer vision and pattern recognition , 770--778

    work page 2016

  11. [11]

    MobileNets: Efficient Convolutional Neural Networks for Mobile Vision Applications

    Howard, A. G.; Zhu, M.; Chen, B.; Kalenichenko, D.; Wang, W.; Weyand, T.; Andreetto, M.; and Adam, H. 2017. Mobilenets: Efficient convolutional neural networks for mobile vision applications. arXiv preprint arXiv:1704.04861

    work page internal anchor Pith review Pith/arXiv arXiv 2017

  12. [12]

    Hu, J.; Goeders, J.; Brisk, P.; Wang, Y.; Luo, G.; and Yu, B. 2019. 2019 DAC system design contest on low power object detection. When Accuracy meets Power: 2019 DAC System Design Contest on Low Power Object Detection

    work page 2019

  13. [13]

    Kara, K., and Alonso, G. 2019. DAC-SDC'19 3rd place winner in FPGA track

    work page 2019

  14. [14]

    Kara, K.; Zhang, C.; and Alonso, G. 2018. DAC-SDC'18 2nd place winner in FPGA track. https://github.com/fpgasystems/spooNN. Accessed: 2019-06-09

    work page 2018

  15. [15]

    Krizhevsky, A.; Sutskever, I.; and Hinton, G. E. 2012. Imagenet classification with deep convolutional neural networks. In Advances in neural information processing systems , 1097--1105

    work page 2012

  16. [16]

    Liu, W.; Anguelov, D.; Erhan, D.; Szegedy, C.; Reed, S.; Fu, C.-Y.; and Berg, A. C. 2016. Ssd: Single shot multibox detector. In European conference on computer vision , 21--37. Springer

    work page 2016

  17. [17]

    Lu, H.; Cai, X.; Zhao, X.; and Wang, Y. 2018. DAC-SDC'18 1st place winner in GPU track. https://github.com/lvhao7896/DAC2018. Accessed: 2019-06-09

    work page 2018

  18. [18]

    Redmon, J.; Divvala, S.; Girshick, R.; and Farhadi, A. 2016. You only look once: Unified, real-time object detection. In Proceedings of the IEEE conference on computer vision and pattern recognition , 779--788

    work page 2016

  19. [19]

    Simonyan, K., and Zisserman, A. 2014. Very deep convolutional networks for large-scale image recognition. arXiv preprint arXiv:1409.1556

    work page internal anchor Pith review Pith/arXiv arXiv 2014

  20. [20]

    Xiong, F.; Yin, S.; Fan, Y.; and Ouyang, P. 2019. DAC-SDC'19 2nd place winner in GPU track

    work page 2019

  21. [21]

    DAC-SDC Low Power Object Detection Challenge for UAV Applications

    Xu, X.; Zhang, X.; Yu, B.; Hu, X. S.; Rowen, C.; Hu, J.; and Shi, Y. 2018. DAC-SDC low power object detection challenge for UAV applications. arXiv preprint arXiv:1809.00110

    work page internal anchor Pith review Pith/arXiv arXiv 2018

  22. [22]

    Zang, C.; Liu, J.; Hao, Y.; Li, S.; Yu, M.; Zhao, Y.; Li, M.; Xue, P.; Qin, X.; Ju, L.; Li, X.; Zhao, M.; and Dai, H. 2018. DAC-SDC'18 3rd place winner in GPU track. https://github.com/xiaoyuuuuu/dac-hdc-2018-object-detection-in-Jetson-TX2. Accessed: 2019-06-09

    work page 2018

  23. [23]

    Zeng, S.; Chen, W.; Huang, T.; Lin, Y.; Meng, W.; Zhu, Z.; and Wang, Y. 2018. DAC-SDC'18 1st place winner in FPGA track. https://github.com/hirayaku/DAC2018-TGIIF. Accessed: 2019-06-09

    work page 2018

  24. [24]

    Zhang, X.; Wang, J.; Zhu, C.; Lin, Y.; Xiong, J.; Hwu, W.-m.; and Chen, D. 2018. DNNBuilder : an automated tool for building high-performance DNN hardware accelerators for FPGAs . In Proceedings of the International Conference on Computer-Aided Design , 56. ACM

    work page 2018

  25. [25]

    Zhang, X.; Hao, C.; Li, Y.; Chen, Y.; Xiong, J.; Hwu, W.-m.; and Chen, D. 2019a. A bi-directional co-design approach to enable deep learning on IoT devices. arXiv preprint arXiv:1905.08369

    work page internal anchor Pith review Pith/arXiv arXiv 1905

  26. [26]

    L.; Hao, C.; Fan, Y.; Li, Y.; Huang, S.; Cheng, B.; Wei, Y.; Huang, T.; Xiong, J.; Shi, H.; Hwu, W.-m.; and Chen, D

    Zhang, X.; Lu, H.; Li, J. L.; Hao, C.; Fan, Y.; Li, Y.; Huang, S.; Cheng, B.; Wei, Y.; Huang, T.; Xiong, J.; Shi, H.; Hwu, W.-m.; and Chen, D. 2019b. DAC-SDC'19 1st place winner in GPU track

    work page

  27. [27]

    Zhao, B.; Zhao, W.; Xia, T.; Chen, F.; Fan, L.; Zong, P.; Wei, Y.; Tu, Z.; Zhao, Z.; Dong, Z.; and Ren, P. 2019. DAC-SDC'19 2nd place winner in FPGA track

    work page 2019