Pith. sign in

REVIEW 4 major objections 7 minor 44 references

Hardware-Aware Deployment of Joint SAR Compression and Despeckling on FPGA

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

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

Pith's one-line read A learned SAR despeckling-compression system can run on an FPGA with no loss in quality and a large energy saving.

desk verdict A genuine first: FPGA deployment of joint SAR despeckling and learned compression, with a surprising ReLU-over-GDN result that deserves a serious referee, though all distortion metrics are anchored to a MERLIN-estimated reference. read the letter →

arxiv 2608.11271 v1 pith:TR4TWPHJ submitted 2026-08-11 eess.IV cs.LG

Cédric Léonard , Francescopaolo Sica , Martin Schulz This is my paper · ORCID
classification eess.IVcs.LG
keywords SARdespecklinglearnedimagecompressionFPGAquantizationrate-distortiononboardprocessinghardware-awaredesign
verification ladder T0 review T1 audit T2 compute T3 formal

The pith

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

The reading

The paper sets out to show that a learned joint SAR despeckling-and-compression model can be reshaped to obey an FPGA accelerator's fixed-point arithmetic and limited operator set without sacrificing rate-distortion performance. It replaces the GDN activations inherited from natural-image compression with plain ReLU and removes output padding from transposed convolutions, finding that both hardware-driven changes preserve or improve quality on SAR data. Comparing four model topologies, it reports that residual blocks add little quality for roughly ten times the compute, and that int8 quantization can raise PSNR while slightly lowering perceptual metrics such as SSIM and edge preservation. The energy payoff is the central practical result: the FPGA prototype consumes 11 to 17 times less energy per patch than a CPU and 2 to 6 times less than a GPU, staying within typical SmallSat power budgets. A full-tile projection suggests a 2.51 GB scene could shrink to about 13 to 132 MB in roughly 4 to 11 minutes, a one-to-two order-of-magnitude reduction in downlink volume.

What carries the argument

The carrying mechanism is the hardware-aware redesign of the DDC autoencoder: the neural transforms g_a, h_a, h_s, and g_s run on the FPGA overlay's int8 deep-learning unit, while entropy coding runs on the embedded ARM CPU. GDN activations are replaced by ReLU, and transposed-convolution kernels are reshaped from (5x5, stride 2, output padding 1) to (4x4, stride 1, output padding 0) to fit the accelerator's operator set. The evaluation axis is rate-distortion measured against a MERLIN-despeckled U-Net estimate of true SAR reflectivity, so the reported PSNR, SSIM, and edge-preservation values are comparative across models and precisions rather than absolute fidelity measures.

What would settle it

Recompute the rate-distortion curves using a reference obtained by multi-looking many independent SLC acquisitions of the same scene or by simulated SAR data with known ground-truth reflectivity, keeping all model and hardware settings fixed; if ReLU's advantage over GDN and the int8-over-float32 PSNR gains shrink or reverse, the despeckled reference bias was the cause.

Watch

Extended reading notes

Core claim

The central claim is that deployment constraints, far from being a nuisance, can guide the architecture toward better SAR compression. For the joint SAR Despeckling and Data Compression (DDC) framework, which trains with the MERLIN self-supervised despeckling loss plus a learned rate term, the authors show that replacing GDN with ReLU improves the rate-distortion trade-off by about 1.67 dB PSNR and 0.26 bpp at the highest evaluated rate, and that removing output padding from transposed convolutions helps slightly. The original residual-plus-hyperprior model therefore performs no worse once made FPGA-compatible. Comparing ResSH, SH, ResFP, and FP, the residual blocks' tenfold compute overhead buys only about 0.75 dB at the high-rate end, making the lightweight factorized-prior model the recommended choice for onboard use. Across all four architectures, int8 post-training quantization scores higher PSNR than the float32 GPU baseline by 0.50 to 1.55 dB at the highest rate, while SSIM drops by about 0.02 and edge preservation drops more, an asymmetry the authors attribute to int8 reconstructions being less bright and therefore closer to the despeckled reference over most of the scene. The energy result is the headline: the FPGA uses 11 to 17 times less energy per patch than the CPU and 2 to 6 times less than the GPU.

Load-bearing premise

All quality comparisons are measured against a despeckled reference that is itself an estimate produced by another network, so if that reference is biased, for example too dark in high-scatterer areas, the ReLU advantage and the int8 PSNR gains could be artifacts of matching a biased target.

Editorial extensions

If this is right

  • A lightweight factorized-prior model with ReLU activations and no residual blocks is sufficient for onboard SAR compression, cutting compute roughly tenfold at near-equal rate-distortion performance.
  • GDN, a standard component in learned compression of natural images, is not needed for SAR speckle statistics; ReLU is both simpler and better aligned with the FPGA's supported operations.
  • int8 quantization can improve pixel-wise PSNR against the despeckled reference while costing perceptual fidelity: SSIM drops slightly and edge preservation drops more, so quality claims should report multiple metrics.
  • The FPGA prototype fits within the 20 to 95 W SmallSat power budget and delivers 11 to 17 times lower energy per patch than a CPU and 2 to 6 times lower than a GPU.
  • Projected to a full TerraSAR-X SLC tile, the method could compress a 2.51 GB scene to 13 to 132 MB in 4.0 to 11.3 minutes, a 19 to 198 times reduction in downlink volume.

Reading between the lines

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

  • A testable extension is to re-evaluate the ReLU-versus-GDN and int8-versus-float32 comparisons against an independent reflectivity reference, for example a multi-look average of many SLC acquisitions; if the advantages shrink or reverse, they are artifacts of matching a biased despeckling estimate.
  • The fact that ReLU beats GDN on SAR suggests that density-modeling activations tuned for natural images may be the wrong inductive bias for coherent imaging, and that other components of natural-image LIC should be re-tested on speckle-dominated data.
  • The log-normalization step executed on the ARM CPU accounts for up to 41 percent of the FPGA's per-patch latency for the cheapest model; moving that operation into the programmable logic is a concrete optimization that the paper identifies only qualitatively.
  • Because the learned latent is speckle-free and compact, task-specific decoders for detection or segmentation could be attached directly to the compressed representation on board, avoiding full decompression before exploitation.
Share X Bluesky LinkedIn Reddit HN

Signed reviews

No signed human review yet.

Editorial analysis

A structured set of objections, weighed in public.

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

Referee Report

4 major / 7 minor

Summary. This manuscript presents the first FPGA deployment of a joint SAR despeckling and compression (DDC) framework, using a ZCU102 board with AMD's Vitis AI DPU. To satisfy the DPU's fixed-point arithmetic and operator constraints, the authors replace GDN activations with ReLU, reshape transposed convolutions, and apply int8 post-training quantization. They evaluate four topologies (FP, ResFP, SH, ResSH) in float32 and int8 across CPU, GPU, and FPGA, reporting RD curves, latency breakdowns, and per-patch energy. Main claims: hardware-driven modifications improve rather than degrade RD (ReLU gains 1.67 dB over GDN at the highest rate); residual blocks add little quality for about 10x the compute; int8 increases PSNR while slightly decreasing SSIM and more strongly decreasing EPD; and the FPGA consumes 11-17x less energy per patch than a CPU and 2-6x less than a GPU. The paper also projects full-tile processing times and downlink savings.

Significance. The paper is a useful systems contribution with reproducible open-source code, multiple training seeds, and error bars on RD curves. If the central claims hold, the finding that deployment-driven simplifications (ReLU, no output padding, int8 PTQ) preserve or improve rate-distortion performance is practically important for onboard SAR, and the cross-platform energy/latency measurements provide a valuable reference for future mission design. The architecture ablation and the explicit identification of residual-block overhead are also informative. However, the evaluation's dependence on a MERLIN-estimated reference and the apparent use of different test sets for FPGA and GPU results are load-bearing concerns that need to be resolved before the conclusions can be fully accepted.

major comments (4)
  1. [IV-C, V-C, Fig. 6] The FPGA int8 results in Fig. 6 appear to be computed on a different test set than the GPU float32 results: Section IV-C states that FPGA results use a 500-image 'representative subset' of the 5,724-image test set, with no selection procedure described, while the GPU baseline is likely evaluated on the complete test set. If this is the case, the reported int8 PSNR gains of +0.50 to +1.55 dB over float32 are not a valid comparison because the two curves are not on the same data. Please evaluate both precisions on the identical images and report the subset selection procedure (e.g., random seed, stratification by scene); this is necessary for the central C2 claim about quantization.
  2. [IV-C, V-A, V-C] All distortion metrics are anchored to a MERLIN-despeckled U-Net reference, which is itself a self-supervised estimate, and the DDC training loss uses the same MERLIN objective (Eq. 5). The paper's caveat that metrics should be read comparatively does not fully address the risk that a biased reference reorders the models, since a reference that is too dark in high-scatterer areas would reward outputs that darken those regions; indeed, Section V-C attributes the int8 PSNR gain to reconstructions that 'match more closely the despeckled reference across the dark majority of the scenes.' To support the surprising ReLU and int8 results, please add an independent validation, for example synthetic speckled scenes with known reflectivity, or a second despeckling reference from a different method, and show that the ranking of architectures and precisions is stable.
  3. [V-C, Fig. 6] The section titled 'The Cost of Quantization' compares GPU float32 with FPGA int8, which confounds the effect of int8 post-training quantization with the change of platform and inference stack (DPU, Vitis AI, CPU-DPU partitioning). To isolate the quantization cost, please report an int8 evaluation of the same models on the GPU (or a float32 reference on the same FPGA toolchain if available), or explicitly reframe the comparison as 'FPGA deployment vs GPU baseline' and avoid claiming a pure quantization effect.
  4. [V-E, Table IV] The central energy-efficiency claim (11-17x less energy per patch than CPU and 2-6x less than GPU) rests on power measurements whose scopes 'differ slightly' (footnote 5). Please provide a detailed measurement methodology for each platform: which components are included in the power draw, how the average is computed, and whether the latency measurements use batch size one. Without this information, the magnitude of the reported energy advantage cannot be verified or reproduced.
minor comments (7)
  1. [IV-E] The transposed-convolution reshaping notation '(5x5,2,1) to (4x4,1,0)' is ambiguous because the definition (k x k, ip, op) does not state the stride; please give the exact layer parameters (kernel, stride, padding, output_padding) for both the original and modified decoders.
  2. [V-A] The sentence reporting the ReLU improvement 'by 1.67 +/- 0.10 dB and 0.26 +/- 0.07 bpp' should specify whether the bpp reduction is at the same lambda or at the same distortion; if at the same lambda, it is a rate reduction at a fixed operating point rather than an independent RD improvement.
  3. [V-E, Table IV] The stated energy ratios '11-17x vs CPU and 2-6x vs GPU' do not match Table IV exactly (e.g., ResFP vs GPU is 1.8x); please reconcile the ranges.
  4. [V-C] The SSIM quantization cost of 'about 0.02 points' should state the rate/lambda at which it is measured and whether it is averaged over the RD curve.
  5. [V-D, Fig. 8] The latency breakdown is informative, but the 'normalize' overhead is executed on an un-vectorized ARM CPU; please state whether this operation could be moved to the FPGA fabric and whether the reported overhead is included in all platform comparisons.
  6. [IV-C] The term 'representative subset' should be replaced with a precise description of how the 500 images were selected, even if the comparison in Fig. 6 is corrected to use the same subset for both precisions.
  7. [Fig. 7] The Hamburg tile is a single high-scatterer scene; consider showing a low-scatterer example to illustrate the average behavior behind the test-set statistics in Fig. 6.

Circularity Check

0 steps flagged · score 0.0 of 10

No circularity: all central claims are empirical comparisons against an acknowledged estimated reference, with no prediction reducing to fitted inputs or load-bearing self-citation.

full rationale

The paper's central findings are empirical measurements, not derivations from fitted constants. The ReLU-versus-GDN result, the residual-block tradeoff, the int8-versus-float32 behavior, and the FPGA energy-efficiency numbers all come from training the DDC variants and measuring rate, PSNR/SSIM/EPD, latency, and energy on CPU, GPU, and FPGA. No quantity is fitted on a subset and then reported as a prediction of a closely related quantity. The MERLIN-despeckled U-Net output used as the evaluation reference is explicitly acknowledged in Section IV-C as 'itself an estimate' and the paper states that 'the resulting metrics should be read comparatively across models rather than as absolute fidelity.' While the DDC training loss and the evaluation reference share a MERLIN lineage, the models are not trained to reproduce that U-Net's output, and the reported differences between architectures and precisions are measured rather than constructed by the evaluation procedure. Self-citations to the DDC framework [5] and to an FPGA review [7] by overlapping authors are background and starting points, but they are not load-bearing for the paper's empirical conclusions. Accordingly, no circular step can be exhibited, and the reference-dependence concern is a validity/correctness caveat rather than a circularity.

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

The paper's quantitative claims rest on a set of modeling and measurement choices inherited from prior work and on evaluation against an estimated reference. The listed hyperparameters and dataset selections shape the reported numbers; the axioms are the unproven, but standard or acknowledged, premises on which the experiments stand.

free parameters (4)
  • RD tradeoff points (lambda grid) = 1, 2, 5, 10, 20, 50, 100, 200, 500, 1000
    Ten rate-distortion tradeoff points used to trace RD curves; standard in LIC, but the specific grid shapes the reported operating points and the 'highest rate' claims.
  • Per-architecture learning rate = 1e-4 (ResSH), 5e-4 (FP/ResFP/SH)
    Selected by grid search over [1e-5, 1e-3]; tuning per architecture is fair, but the difference could influence the architecture comparison.
  • FPGA test subset size = 500 of 5,724 test images
    FPGA results evaluated on a subset described as representative; selection method not specified, affecting the FPGA vs GPU comparison.
  • Patch overlap for full-tile projection = 16 px
    Chosen to avoid edge artifacts in the end-to-end onboard projection; affects the tile latency and energy estimate.
assumptions (4)
  • domain assumption MERLIN-generated despeckled images are a valid reference for evaluating reconstruction fidelity.
    The paper states this reference is itself an estimate and metrics should be read comparatively; all PSNR, SSIM, and EPD values are anchored to this internal proxy rather than to true reflectivity (Section IV-C, V-A).
  • domain assumption The ZCU102 with Vitis AI DPU is a representative proxy for spaceborne FPGA constraints.
    The board is a commercial evaluation kit, not radiation-hardened space-grade hardware; the paper uses it to prototype the deployment workflow (Section II-C, IV-D).
  • domain assumption The DDC training objective (MERLIN loss plus LIC rate) yields useful co-despeckling and compression.
    Adopted without re-derivation from Amao-Oliva et al. [5]; the paper's conclusions about ReLU vs GDN are conditional on this training framework being effective.
  • domain assumption The Noise2Noise premise that Re^2 and Im^2 of SLC data are independent realizations of the same process.
    Inherited from MERLIN [26]; it underpins the self-supervised training strategy used for all DDC models.

how reviews work

0 comments
Cite this review

Pith. "Pith review of Hardware-Aware Deployment of Joint SAR Compression and Despeckling on FPGA." pith.science (2026). https://pith.science/paper/TR4TWPHJ

@misc{pith2026260811271,
  author       = {Pith},
  title        = {Pith review of: Hardware-Aware Deployment of Joint SAR Compression and Despeckling on FPGA},
  year         = {2026},
  howpublished = {\url{https://pith.science/paper/TR4TWPHJ}},
  note         = {Machine review of arXiv:2608.11271}
}
read the original abstract

Next-generation Synthetic Aperture Radar (SAR) missions will generate data far faster than they can downlink, making onboard data reduction essential for near-real-time Earth observation. Learned Image Compression (LIC) offers better rate-distortion performance than handcrafted codecs used operationally today, and recent work shows that simultaneously despeckling and compressing SAR imagery enables better representation capacity while unlocking higher compression rates. These methods, however, have yet to be confronted with the strict power, compute, and operational constraints of spaceborne systems. In this work, we bridge this gap by deploying a joint SAR Despeckling and Data Compression (DDC) framework on an embedded ZCU102 FPGA-based platform, introducing model adaptations that respect the accelerator's fixed-point arithmetic and limited set of supported operations. We evaluate four model topologies across precision levels and across CPU, GPU, and FPGA platforms, revealing several findings with direct design implications. We find that replacing conventional GDN activation functions with plain ReLU improves quality on SAR, suggesting that design principles established for compression of natural images do not necessarily transfer to SAR imagery. In addition, we demonstrate that residual blocks offer little representational benefit for ten times the compute, and show that the FPGA is the most energy-efficient of the platforms tested. Together, these results set a functioning edge deployment workflow and an evidence-based starting point for onboard SAR compression. The code is available at https://github.com/CedricLeon/SAR_DDC_FPGA.

Figures

Figures reproduced from arXiv: 2608.11271 by the authors.

Figure 1
Figure 1. Architecture diagrams of the Factorized Prior ( [PITH_FULL_IMAGE:figures/full_fig_p003_1.png] view at source ↗
Figure 2
Figure 2. DDC encoder-side architecture. (a) Main analysis transform [PITH_FULL_IMAGE:figures/full_fig_p005_2.png] view at source ↗
Figure 3
Figure 3. DDC inference dataflow. The left of the figure, (blue background) shows the [PITH_FULL_IMAGE:figures/full_fig_p006_3.png] view at source ↗
Figures from the paper (5 more)
Figure 4
Figure 4. Figure 4: FPGA-aware modifications of the ResSH architecture, in float32 precision. Each RD-curve has varying activation functions (ReLU / GDN, cf [PITH_FULL_IMAGE:figures/full_fig_p007_4.png]
Figure 5
Figure 5. Figure 5: RD performance of four model topologies. All architectures are [PITH_FULL_IMAGE:figures/full_fig_p007_5.png]
Figure 6
Figure 6. Figure 6: RD-curves of the four architectures across two different hardware platforms and precision GPU ( [PITH_FULL_IMAGE:figures/full_fig_p008_6.png]
Figure 7
Figure 7. Figure 7: Comparison of reconstructions over Hamburg city-center ( [PITH_FULL_IMAGE:figures/full_fig_p009_7.png]
Figure 8
Figure 8. Figure 8: Breakdown of the latencies of each architecture to compress one patch [PITH_FULL_IMAGE:figures/full_fig_p009_8.png]

Discussion (0). Continue with ORCID to comment.

Reference graph

Works this paper leans on

44 extracted references · 35 canonical work pages

  1. [1]

    Block adaptive quantization,

    D. C. Lancashire, B. A. F. Barnes, and S. J. Udall, “Block adaptive quantization,” US Patent US6 255 987B1, Jul., 2001. [Online]. Available: https://patents.google.com/patent/US6255987B1/en

    work page 2001

  2. [2]

    Flexible Dynamic Block Adaptive Quantization for Sentinel-1 SAR Missions,

    E. Attema, C. Cafforio, M. Gottwald, P. Guccione, A. Monti Guarnieri, F. Rocca, and P. Snoeij, “Flexible Dynamic Block Adaptive Quantization for Sentinel-1 SAR Missions,”IEEE Geoscience and Remote Sensing Letters, vol. 7, no. 4, pp. 766–770, Oct. 2010. [Online]. Available: https://ieeexplore.ieee.org/abstract/document/5473136

    work page arXiv 2010

  3. [3]

    End-to-end Optimized Image Compression,

    J. Ball ´e, V . Laparra, and E. P. Simoncelli, “End-to-end Optimized Image Compression,” Mar. 2017, arXiv:1611.01704 [cs, math]. [Online]. Available: http://arxiv.org/abs/1611.01704

    arXiv 2017

  4. [4]

    Variational image compression with a scale hyperprior,

    J. Ball ´e, D. Minnen, S. Singh, S. J. Hwang, and N. Johnston, “Variational image compression with a scale hyperprior,” inInternational Conference on Learning Representations 2018, Feb. 2018. [Online]. Available: https://openreview.net/forum?id=rkcQFMZRb

    work page 2018

  5. [5]

    Joint compression and despeckling by SAR representation learning,

    J. Amao-Oliva, N. Foix-Colonier, and F. Sica, “Joint compression and despeckling by SAR representation learning,”ISPRS Journal of Photogrammetry and Remote Sensing, vol. 220, pp. 524–534, 2025. [Online]. Available: https://www.sciencedirect.com/science/article/pii/ S0924271624004866

    work page 2025

  6. [6]

    Small Spacecraft Technology State of the Art,

    NASA, Washington, DC, USA, Tech. Rep., “Small Spacecraft Technology State of the Art,”NASA/TP-2015-216648/REV, 2015. [Online]. Available: https://sst-soa.arc.nasa.gov

    work page 2015

  7. [7]

    FPGA-Enabled Machine Learning Applications in Earth Observation: A Systematic Review,

    C. Leonard, D. Stober, and M. Schulz, “FPGA-Enabled Machine Learning Applications in Earth Observation: A Systematic Review,” ACM Comput. Surv., vol. 58, no. 11, pp. 283:1–283:36, Apr. 2026. [Online]. Available: https://dl.acm.org/doi/10.1145/3800686

    work page doi:10.1145/3800686 2026

  8. [8]

    AMD Vitis™ AI Software,

    AMD, “AMD Vitis™ AI Software,” 2019. [Online]. Available: https://www.amd.com/en/products/software/vitis-ai.html

    work page 2019

Show all 44 references
  1. [9]

    Reducing the Dimensionality of Data with Neural Networks,

    G. E. Hinton and R. R. Salakhutdinov, “Reducing the Dimensionality of Data with Neural Networks,”Science, vol. 313, no. 5786, pp. 504–507, Jul. 2006, publisher: American Association for the Advancement of Science. [Online]. Available: https://www.science.org/doi/full/10.1126/ ...

    2006

  2. [10]

    JPEG2000: Image Compression Fundamentals, Standards and Practice,

    M. Rabbani, “JPEG2000: Image Compression Fundamentals, Standards and Practice,”Journal of Electronic Imaging, vol. 11, no. 2, p. 286,

  3. [11]

    Joint Autoregressive and Hierarchical Priors for Learned Image Compression,

    D. Minnen, J. Ball ´e, and G. D. Toderici, “Joint Autoregressive and Hierarchical Priors for Learned Image Compression,” inAdvances in Neural Information Processing Systems, vol. 31. Curran Associates, Inc., 2018. [Online]. Available: https://proceedings.neurips.cc/paper/ 2018...

    2018

  4. [12]

    Power Consumption of GPUs from a Software Perspective,

    C. Collange, D. Defour, and A. Tisserand, “Power Consumption of GPUs from a Software Perspective,” inComputational Science – ICCS 2009, G. Allen, J. Nabrzyski, E. Seidel, G. D. van Albada, J. Dongarra, and P. M. A. Sloot, Eds. Berlin, Heidelberg: Springer, 2009, pp. 914–923

    2009

  5. [13]

    The Final Frontier: Deep Learning in Space,

    V . Kothari, E. Liberis, and N. D. Lane, “The Final Frontier: Deep Learning in Space,” inProceedings of the 21st International Workshop on Mobile Computing Systems and Applications. Austin TX USA: ACM, Mar. 2020, pp. 45–49. [Online]. Available: https://dl.acm.org/doi/10.1145/3...

    2020

  6. [14]

    NVIDIA Jetson Nano Module,

    NVIDIA, “NVIDIA Jetson Nano Module,” 2019. [Online]. Available: https://developer.nvidia.com/embedded/jetson-nano

    2019

  7. [15]

    Machine Learning in Space: Surveying the Robustness of On-Board ML Models to Radiation,

    K. Lange, F. Fontana, F. Rossi, M. Varile, and G. Apruzzese, “Machine Learning in Space: Surveying the Robustness of On-Board ML Models to Radiation,” in2024 IEEE Space Computing Conference (SCC). Mountain View, CA, USA: IEEE, Jul. 2024, pp. 51–64. [Online]. Available: https:/...

    2024

  8. [16]

    Intel® Movidius™ Myriad™ 2 Vision Processing Unit 1GB - Product Specifications,

    Intel, “Intel® Movidius™ Myriad™ 2 Vision Processing Unit 1GB - Product Specifications,” Tech. Rep., 2016. [Online]. Avail- able: https://www.intel.com/content/www/us/en/products/sku/204771/ intel-movidius-myriad-2-vision-processing-unit-1gb/specifications.html

    2016

  9. [17]

    An FPGA-based hardware accelerator for cnns inference on board satellites: Benchmarking with myriad 2-based solution for the CloudScout case study,

    E. Rapuano, G. Meoni, T. Pacini, G. Dinelli, G. Furano, G. Giuffrida, and L. Fanucci, “An FPGA-based hardware accelerator for cnns inference on board satellites: Benchmarking with myriad 2-based solution for the CloudScout case study,”REMOTE SENSING, vol. 13, no. 8, Apr. 2021,...

    2021

  10. [18]

    FPGA vs. ASIC for low power applications,

    A. Amara, F. Amiel, and T. Ea, “FPGA vs. ASIC for low power applications,”Microelectronics Journal, vol. 37, no. 8, pp. 669–677, Aug. 2006. [Online]. Available: https://www.sciencedirect.com/science/ article/pii/S0026269205003927

    2006

  11. [19]

    Mitigation of Radiation Effects in SRAM-Based FPGAs for Space Applications,

    F. Siegle, T. Vladimirova, J. Ilstad, and O. Emam, “Mitigation of Radiation Effects in SRAM-Based FPGAs for Space Applications,” ACM Comput. Surv., vol. 47, no. 2, pp. 37:1–37:34, Jan. 2015. [Online]. Available: https://dl.acm.org/doi/10.1145/2671181

    2015 doi

  12. [20]

    Onboard Processing With Hybrid and Reconfigurable Computing on Small Satellites,

    A. D. George and C. M. Wilson, “Onboard Processing With Hybrid and Reconfigurable Computing on Small Satellites,”Proceedings of the IEEE, vol. 106, no. 3, pp. 458–470, Mar. 2018, conference Name: Proceedings of the IEEE. [Online]. Available: https://ieeexplore.ieee. org/abstra...

    2018

  13. [21]

    FINN: A Framework for Fast, Scalable Binarized Neural Network Inference,

    Y . Umuroglu, N. J. Fraser, G. Gambardella, M. Blott, P. Leong, M. Jahre, and K. Vissers, “FINN: A Framework for Fast, Scalable Binarized Neural Network Inference,” inProceedings of the 2017 ACM/SIGDA International Symposium on Field-Programmable Gate Arrays. Monterey Californ...

    2017

  14. [22]

    On the use of JPEG2000 for SAR raw data compression,

    R. M. Asiyabi, A. Anghel, A. Focsa, M. Datcu, M. Martone, P. Rizzoli, and E. Imbembo, “On the use of JPEG2000 for SAR raw data compression,” inEUSAR 2024; 15th European Conference on Synthetic Aperture Radar, Apr. 2024, pp. 249–253. [Online]. Available: https://ieeexplore.ieee...

    2024

  15. [23]

    AI-BAQ: Deep Learning for Adaptive SAR Raw Data Quantization,

    N. Gollin, M. Martone, E. Imbembo, M. Ghiglione, S. Knoll, G. Krieger, and P. Rizzoli, “AI-BAQ: Deep Learning for Adaptive SAR Raw Data Quantization,”IEEE Transactions on Geoscience and Remote Sensing, vol. 63, pp. 1–20, 2025. [Online]. Available: https://ieeexplore.ieee.org/d...

    2025

  16. [24]

    Complex-Valued Autoencoder for Multi-Polarization SLC SAR Data Compression with Side Information,

    R. M. Asiyabi, A. Anghel, P. Rizzoli, M. Martone, and M. Datcu, “Complex-Valued Autoencoder for Multi-Polarization SLC SAR Data Compression with Side Information,” inIGARSS 2023 - 2023 IEEE International Geoscience and Remote Sensing Symposium, Jul. 2023, pp. 1787–1790, iSSN: ...

    2023

  17. [25]

    Optimizing Image Compression via Joint Learning with Denoising,

    K. L. Cheng, Y . Xie, and Q. Chen, “Optimizing Image Compression via Joint Learning with Denoising,” inComputer Vision – ECCV 2022: 17th European Conference, Tel Aviv, Israel, October 23–27, 2022, Proceedings, Part XIX. Berlin, Heidelberg: Springer-Verlag, Oct. 2022, pp. 56–73...

    2022

  18. [26]

    As if by magic: self- supervised training of deep despeckling networks with MERLIN,

    E. Dalsasso, L. Denis, and F. Tupin, “As if by magic: self- supervised training of deep despeckling networks with MERLIN,” IEEE Transactions on Geoscience and Remote Sensing, vol. 60, pp. 1–13, 2022, arXiv:2110.13148 [cs, eess]. [Online]. Available: http://arxiv.org/abs/2110.13148

    2022 arXiv

  19. [27]

    Noise2Noise: Learning Image Restoration without Clean Data,

    J. Lehtinen, J. Munkberg, J. Hasselgren, S. Laine, T. Karras, M. Aittala, and T. Aila, “Noise2Noise: Learning Image Restoration without Clean Data,” Oct. 2018, arXiv:1803.04189 [cs]. [Online]. Available: http://arxiv.org/abs/1803.04189

    2018 arXiv

  20. [28]

    FPGA Codec System of Learned Image Compression With Algorithm-Architecture Co-Optimization,

    H. Sun, Q. Yi, and M. Fujita, “FPGA Codec System of Learned Image Compression With Algorithm-Architecture Co-Optimization,” IEEE Journal on Emerging and Selected Topics in Circuits and Systems, vol. 14, no. 2, pp. 334–347, Jun. 2024. [Online]. Available: https://ieeexplore.iee...

    2024

  21. [29]

    Exploring Efficient Hardware Accelerator for Learning-Based Image Compression,

    C. Chen, H. Zhang, K. Guo, X. Yu, W. Qiu, Z. Qi, and H. Guan, “Exploring Efficient Hardware Accelerator for Learning-Based Image Compression,”IEEE Transactions on Computer-Aided Design of Integrated Circuits and Systems, vol. 44, no. 6, pp. 2204–2217, Jun. 2025. [Online]. Avai...

    2025

  22. [30]

    Design and implementation of CNN- based Custom Net Architecture with improved inference time for Real- time Remote Sensing Application,

    G. Upadhyay, S. Ghosal, S. Kart, K. Jain, S. Shantala, J. LalitKrushna, S. Srividhya, and S. Balwantrao, “Design and implementation of CNN- based Custom Net Architecture with improved inference time for Real- time Remote Sensing Application,” in2024 IEEE SPACE, AEROSPACE AND D...

    2024

  23. [31]

    A methodology for evaluating and analyzing FPGA-accelerated, deep-learning applications for onboard space pro- cessing,

    S. Sabogal and A. George, “A methodology for evaluating and analyzing FPGA-accelerated, deep-learning applications for onboard space pro- cessing,” in2021 ieee space computing conference (scc), 2021, pp. 143– 154, type: Proceedings Paper tex.affiliation: Sabogal, S (Correspond...

    2021

  24. [32]

    An Improved Lightweight Deep Learning Model and Implementation for Track Fastener Defect Detection with Unmanned Aerial Vehicles,

    Q. Yu, A. Liu, X. Yang, and W. Diao, “An Improved Lightweight Deep Learning Model and Implementation for Track Fastener Defect Detection with Unmanned Aerial Vehicles,”ELECTRONICS, vol. 13, no. 9, May 2024

    2024

  25. [33]

    An Edge and Trustworthy AI UA V System With Self-Adaptivity and Hyperspectral Imaging for Air Quality Monitoring,

    C. Huang, W. Chen, Y . Chang, and K. Wu, “An Edge and Trustworthy AI UA V System With Self-Adaptivity and Hyperspectral Imaging for Air Quality Monitoring,”IEEE INTERNET OF THINGS JOURNAL, vol. 11, no. 20, pp. 32 572–32 584, Oct. 2024

    2024

  26. [34]

    FPX-NIC: An FPGA-Accelerated 4K Ultra-High-Definition Neural Video Coding System,

    C. Jia, X. Hang, S. Wang, Y . Wu, S. Ma, and W. Gao, “FPX-NIC: An FPGA-Accelerated 4K Ultra-High-Definition Neural Video Coding System,”IEEE Transactions on Circuits and Systems for Video Technology, vol. 32, no. 9, pp. 6385–6399, Sep. 2022. [Online]. Available: https://ieeexp...

    2022

  27. [35]

    Learned Image Codec on FPGA: Algorithm, Architecture and System Design,

    H. Sun, J. Wang, S. Liu, S. Kimura, and M. Fujita, “Learned Image Codec on FPGA: Algorithm, Architecture and System Design,” in Proceedings of the 30th Asia and South Pacific Design Automation Conference, ser. ASPDAC ’25. New York, NY , USA: Association for Computing Machinery...

    2025

  28. [36]

    Lightweight Embedded FPGA Deployment of Learned Image Compression with Knowledge Distillation and Hybrid Quantization,

    A. Mazouz, S. Chaudhuri, M. Cagnanzzo, M. Mitrea, E. Tartaglione, and A. Fiandrotti, “Lightweight Embedded FPGA Deployment of Learned Image Compression with Knowledge Distillation and Hybrid Quantization,” Mar. 2025, arXiv:2503.04832 [cs]. [Online]. Available: http://arxiv.org...

    2025 arXiv

  29. [37]

    Density Modeling of Images using a Generalized Normalization Transformation,

    J. Ball ´e, V . Laparra, and E. P. Simoncelli, “Density Modeling of Images using a Generalized Normalization Transformation,” Feb. 2016, arXiv:1511.06281 [cs]. [Online]. Available: http://arxiv.org/abs/1511. 06281

    2016 arXiv

  30. [38]

    U-Net: Convolutional Net- works for Biomedical Image Segmentation,

    O. Ronneberger, P. Fischer, and T. Brox, “U-Net: Convolutional Net- works for Biomedical Image Segmentation,” inMedical Image Com- puting and Computer-Assisted Intervention – MICCAI 2015, N. Navab, J. Hornegger, W. M. Wells, and A. F. Frangi, Eds. Cham: Springer International ...

    2015

  31. [39]

    CompressAI: a PyTorch library and evaluation platform for end-to-end compression research,

    J. B ´egaint, F. Racap ´e, S. Feltman, and A. Pushparaja, “CompressAI: a PyTorch library and evaluation platform for end-to-end compression research,” Nov. 2020, arXiv:2011.03029 [cs, eess]. [Online]. Available: http://arxiv.org/abs/2011.03029

    2020 arXiv

  32. [40]

    Adam: A Method for Stochastic Optimization,

    D. P. Kingma and J. Ba, “Adam: A Method for Stochastic Optimization,” Jan. 2017, arXiv:1412.6980 [cs]. [Online]. Available: http://arxiv.org/abs/1412.6980

    2017 arXiv

  33. [41]

    A Survey of Quantization Methods for Efficient Neural Network Inference,

    A. Gholami, S. Kim, Z. Dong, Z. Yao, M. W. Mahoney, and K. Keutzer, “A Survey of Quantization Methods for Efficient Neural Network Inference,” Jun. 2021, arXiv:2103.13630 [cs]. [Online]. Available: http://arxiv.org/abs/2103.13630

    2021 arXiv

  34. [42]

    Real Time Floating Point SAR Focusing on FPGA,

    S. Mandapati, U. Balss, and H. Breit, “Real Time Floating Point SAR Focusing on FPGA,” inEUSAR 2024; 15th European Conference on Synthetic Aperture Radar, Apr. 2024, pp. 60–65. [Online]. Available: https://ieeexplore.ieee.org/document/10659675

    2024

  35. [43]

    SoC FPGA Acceleration for Semantic Segmentation of Clouds in Satellite Images,

    E. A. Papatheofanous, P. Tziolos, V . Kalekis, T. Amrou, G. Konstantoulakis, G. Venitourakis, and D. Reisis, “SoC FPGA Acceleration for Semantic Segmentation of Clouds in Satellite Images,” in2022 IFIP/IEEE 30th International Conference on Very Large Scale Integration (VLSI-So...

    2022

  36. [2002]

    Available: https://doi.org/10.1117/1.1469618

    [Online]. Available: https://doi.org/10.1117/1.1469618

    doi

Pith tools

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