Easz: An Agile Transformer-based Image Compression Framework for Resource-constrained IoTs

Chun Jason Xue; Hong Xu; Jingzong Li; Jun Wang; Nan Guan; Tei-Wei Kuo; Yu Mao

REVIEW 4 major objections 4 minor 34 references

Easz: An Agile Transformer-based Image Compression Framework for Resource-constrained IoTs

T0 review · 4 major / 4 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 Easz erases image patches on the edge and rebuilds them on the server

desk verdict The edge-efficiency story is credible and worth engaging; the reconstruction-quality claim is confounded by Kodak fine-tuning and no-reference metrics. read the letter →

arxiv 2505.01742 v2 pith:FVUOWU6L submitted 2025-05-03 eess.IV cs.LG

Yu Mao , Jingzong Li , Jun Wang , Hong Xu , Tei-Wei Kuo , Nan Guan , Chun Jason Xue This is my paper · ORCID

classification eess.IVcs.LG

keywords imagecompressionedgecomputingtransformererase-and-squeezepatcherasingperceptualqualityIoTdevicesconditionalsampling

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

Easz is an image coding framework that moves the computationally heavy part of neural compression off the edge device and onto the server. On the edge, it applies a row-based conditional sampler to erase a chosen fraction of image sub-patches, squeezes the survivors together, and compresses the result with any existing codec such as JPEG or BPG. The server then runs a lightweight transformer-based encoder-decoder that reconstructs the erased content, restoring the full image. The paper argues that this design avoids the multi-second encoding and model-switching costs of learned compressors on devices like the Jetson TX2, while improving perceptual quality at matched bitrates. If correct, Easz offers a practical way to get learned-quality image compression onto resource-constrained IoT devices.

What carries the argument

The load-bearing components are (1) the row-based conditional sampler that generates the binary erase mask, choosing for each row column positions at least δ away from earlier picks in the same row and at least Δ away from the previous row's picks, which yields flexible, well-distributed erasure with a single reconstruction model; (2) the two-stage patchify process, which splits the image into n×n patches and then into b×b sub-patches and runs attention within each patch, reducing the theoretical attention complexity from O((hw)^2) to O(hw·$n^{2}$/$b^{4}$), making pixel-level reconstruction feasible; and (3) the lightweight 8.7MB transformer encoder-decoder, trained with L1 plus LPIPS loss, that takes embeddings of un-erased sub-patches plus zero vectors at erased positions and outputs the restored sub-patches.

What would settle it

Take a held-out test set the model has never seen, fine-tune on a separate training set, then compute PSNR and SSIM between the original and Easz-reconstructed images at the same bitrates as JPEG/BPG; if reference-based metrics show Easz performs worse than the baselines while the no-reference metrics show it performing better, the claimed quality improvement would be shown to be an artifact of the chosen metrics. Alternatively, decoding with the model trained without the Kodak fine-tuning step and comparing on Kodak would expose whether test-set tuning drives the reported gains.

Watch

Extended reading notes

Core claim

Easz's central claim is that erasing a deliberately spaced subset of image sub-patches at the sender, squeezing the survivors, and reconstructing the erased content on the receiver with a lightweight transformer is enough to match or beat both traditional codecs (JPEG, BPG) and learned compression models (MBT, Cheng-anchor) in perceptual quality at the same bitrate, while cutting the sender's computation to nearly zero. The paper argues that the row-based conditional sampler — which keeps erased blocks non-adjacent both along a row and across neighboring rows — preserves enough local structure that the receiver's model can regenerate the missing detail, and that because erase ratio is just a parameter of the sampler, one trained model serves all compression levels without model switching. In evaluation, Easz improves the perceptual scores of every baseline it is prepended to, and the end-to-end latency on a Jetson TX2 to server testbed is about 89% lower than the neural baselines, with the erase-and-squeeze step taking 0.7% of total latency.

Load-bearing premise

The claimed quality advantage rests on the load-bearing assumption that the chosen no-reference metrics (BRISQUE, Pi, Tres) are a trustworthy measure of reconstruction quality, because the paper deliberately avoids PSNR/SSIM for its compression comparisons — erasing content lowers those reference metrics by construction — and because the reconstruction model is fine-tuned on the Kodak test set before Kodak results are reported.

Editorial extensions

If this is right

Edge devices with no GPU can perform compression-level switching instantly, since changing the erase ratio is just changing a parameter of the sampler, not loading a new model.
Existing deployed compressors (JPEG, BPG, and learned codecs) get a perceptual quality boost at roughly constant bitrate when Easz is prepended, so the framework can be adopted without replacing the codec stack.
The server-side reconstruction model at 8.7MB is small enough to run on modest server GPUs, and the paper's latency breakdown suggests most of its end-to-end latency is reconstruction, which would shrink on faster datacenter hardware.
Because the erase mask is a tiny binary map (128 bytes at 32×32), the sender can adapt bitrate per image or per region with essentially no extra signaling cost.

Reading between the lines

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

The row-based sampler's structure suggests a direct connection to coding theory: it builds a constrained sampling pattern that preserves local correlations; one could test whether optimizing the δ and Δ spacings with a rate-distortion objective rather than fixed thresholds yields further quality or bitrate gains.
The paper's quality comparisons rely on no-reference perceptual metrics; a natural extension is to pair Easz with reference-based or task-based evaluation, such as downstream object detection on reconstructed images, to see whether the perceptual gains translate to machine-vision performance.
Since the reconstruction is trained on CIFAR-10 and fine-tuned on Kodak, the framework's performance on out-of-distribution imagery, such as medical, satellite, or 4K video frames, is an open question; evaluating across diverse domains would test how general the generative prior is.
The two-stage patchify complexity reduction is independent of the erasure scheme and could be reused as a generic way to make pixel-level transformers tractable, not only in compression but in any dense prediction task.

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.

Desk Editor's Note

read the letter

Colleague,

Here is my read of Easz. The new piece is a generalized row-based erase sampler for edge-to-server image compression, plus a lightweight two-stage patchified transformer that reconstructs erased blocks on the server. The special cases reduce to known diagonal masks and 2x super-resolution, so the novelty is incremental, but the generalization to arbitrary erase ratios with one model is real and useful for IoT.

What the paper does well is measure the edge-side cost. The Jetson TX2 numbers are direct, repeated 24 times, and the comparison against MBT and Cheng-anchor is on point: erase-and-squeeze costs almost nothing on the edge, and the power and memory savings are large. The server-side reconstruction being heavier is acknowledged up front. That part of the paper should survive review.

The soft spot is the quality evaluation, and it is load-bearing. Section IV-D shows the reconstruction model was fine-tuned on Kodak ('Fig. 7d' reports MSE during Kodak fine-tuning). Table II then reports Kodak results for the '+Proposed' rows. Unless the fine-tuning split is disjoint from the evaluation split, those rows are test-set adaptation, not independent results. The paper does not say a separate split was used. That is a real confound and needs to be fixed before the quality claim is taken seriously.

Relatedly, the paper avoids PSNR/SSIM for codec comparisons and uses Brisque, Pi, and Tres instead. As the authors note, erasure lowers reference-based metrics, but no-reference metrics can reward the smoothing and artifact removal that erasure plus transformer reconstruction provides. That means the perceptual-quality wins in Table II and Fig. 8 could partly reflect noise removal rather than real detail preservation. Human-rated or region-based reference metrics would address this.

One smaller issue: the complexity worked example in Section III-B gives 1,048,576 calculations after patchification for n=32, b=4, but plugging their own formula O(hw*n^2/b^4) yields 262,144. The text says a 4096x reduction from the original 4.29e9, which matches 1,048,576, but the formula and example don't agree. It's a fixable arithmetic inconsistency, but it needs correcting.

There is also no variance reported for the Table II numbers, and the claim of compatibility with all existing compressors is tested on only four. Minor.

Overall: the efficiency contribution is credible and likely useful for edge vision systems. The reconstruction-quality advantage is not yet established. I'd send this to peer review with the expectation of major revision on the evaluation: separate the fine-tuning split, add human or reference-based quality checks, fix the complexity arithmetic. The systems-side contribution deserves referee time.

Referee Report

4 major / 4 minor

Summary. The paper proposes Easz, an edge-server image compression framework in which the edge device performs only a cheap erase-and-squeeze operation on image patches using a row-based conditional sampler, while a server-side lightweight transformer reconstructs the erased content. The framework is designed to wrap existing codecs such as JPEG, BPG, and learned compressors, and is evaluated on a Jetson TX2 plus RTX 2080Ti testbed. The authors claim that Easz reduces edge latency, power, and memory, provides flexible compression-level adjustment without model switching, and improves reconstruction quality relative to both super-resolution methods and standard codecs.

Significance. The core architectural idea is practically motivated and potentially valuable: moving essentially all neural computation to the server while keeping the edge operation as a simple masking and repacking step is a clean way to address the resource constraints of IoT devices. The measured edge latency, power, and memory advantages reported in Fig. 6 are concrete and credible, and the model-size reduction to 8.7MB is a useful contribution. The flexibility of the row-based sampler is also a strength. However, the paper's headline claim of improved reconstruction quality is not currently established because the evaluation procedure leaks test-set information and relies on no-reference metrics that are acknowledged to favor the erase-and-reconstruct pipeline. The significance of the paper therefore depends on whether the evaluation can be repaired.

major comments (4)

[Sec. IV-D, Fig. 7d, Table II] The reconstruction model is fine-tuned on the Kodak dataset (Fig. 7d) and then Table II reports the headline quality comparisons on the same Kodak dataset. This is a test-set leakage: the Kodak rows in Table II do not measure generalization to unseen images. To support the quality claim, the authors must either fine-tune only on a held-out portion of Kodak and evaluate on a disjoint test portion, or fine-tune on a different dataset and report results on Kodak and CLIC without Kodak fine-tuning. Without this change, the Kodak quality comparisons in Table II are not independent evidence.
[Sec. IV-A, Tables I-II, Fig. 8] The paper explicitly avoids PSNR and SSIM when comparing with compression baselines because 'removing content would generally impact reference-based metrics,' and instead uses the no-reference metrics Brisque, Pi, and Tres. No-reference metrics can reward smoothing and artifact removal even when true detail is lost, so the reported quality improvements may partly reflect the smoothing effect of erasure and reconstruction rather than faithful regeneration. The authors should report reference-based metrics (PSNR, SSIM, and ideally LPIPS) for all methods at matched bitrates, and should also provide a perceptual user study or other validation that the no-reference gains correspond to genuine quality improvement. This is load-bearing because the conclusion that Easz 'enhances reconstruction quality' rests on these metrics.
[Sec. III-B, complexity analysis] The complexity derivation contains a dimensional error. The text states that after two-stage patchification 'there will be (hw)^2/n^2 × n^2/b^2 sub-patches,' but the total number of sub-patches in an h×w image should be hw/b^2, not (hw)^2/b^2. The subsequent numeric example for n=32, b=4 gives 1,048,576 required attention operations, whereas the correct value is 262,144. The qualitative conclusion that patchification reduces attention complexity is correct, but the equations and the claimed '4096 times' reduction must be corrected and recomputed.
[Sec. IV-C, Table I] The comparison with super-resolution methods in Table I does not state the reduction ratio, scale factor, or bitrate at which each method is evaluated. Without matched experimental conditions, the reported PSNR and SSIM differences cannot be interpreted. The authors should specify the erase ratio or scale factor for Easz and for each SR baseline, and ideally report bitrate so that the comparison is rate-constrained.

minor comments (4)

[Sec. IV-D, Fig. 7d] The caption of Fig. 7d says 'MSE,' but the y-axis is labeled 'Loss value' and the training loss in Eq. (2) is L1 plus LPIPS, not MSE. Please align the caption, axis label, and loss definition.
[Sec. IV-A] The phrase 'non-reference perceptual metrics' should be 'no-reference perceptual metrics' throughout the section.
[Reference [26]] Reference [26] is cited as 'K. Alex and H. Geoffrey'; the standard citation is A. Krizhevsky and G. Hinton, 'Learning multiple layers of features from tiny images,' 2009. Please correct the author names.
[Sec. III-A, Eq. (1)] The inter-row constraint uses T both as the number of sampled columns per row and as the index of the last sample in the previous row. Please clarify the notation, for example by writing the previous row's samples as a set and requiring separation from all of them.

Circularity Check

1 steps flagged · score 6.0 of 10

Kodak quality comparisons are confounded: the reconstruction model is fine-tuned on Kodak (Sec. IV-D) and then evaluated on Kodak (Table II), with reference-based metrics avoided in codec comparisons.

fitted input called prediction [Sec. IV-D 'Effectiveness of fine-tuning'; Table II 'Compression Performance Enhancement on Kodak Dataset and Clic Dataset']
"We tested if fine-tuning our pretrained model with the Kodak dataset would be beneficial and found that it indeed improves performance by reducing losses across different patch sizes (1×1, 2×2, and 4×4), as shown in Fig. 7d. ... Table II: Compression Performance Enhancement on Kodak Dataset and Clic Dataset."

The quality claim on Kodak reduces to the fine-tuning fit: the model is fine-tuned on Kodak (Sec. IV-D) and then scored on Kodak (Table II), without any statement that the Kodak evaluation uses the CIFAR-only pretrained model. Thus the reported improvement in Brisque/Pi/Tres on Kodak is, at least in part, performance on training data, not a prediction on unseen images. The paper's decision to avoid PSNR/SSIM in codec comparisons (Sec. IV-A) makes the no-reference metrics the sole quality evidence, and those metrics can improve from smoothing/artifact removal rather than from true recovery of erased content. The latency and adaptability results remain independently measured, so the circularity is limited to the reconstruction-quality pillar.

full rationale

The edge-latency, power, memory, and adaptability claims are supported by direct testbed measurements and mask-flexibility demonstration; they are not circular. The reconstruction-quality advantage, however, is not self-contained. Section IV-D fine-tunes the pretrained model on the Kodak dataset and reports loss reduction on Kodak; Table II then reports Kodak quality numbers. Unless the Kodak evaluation is explicitly run with the CIFAR-only pretrained model (the paper does not say so), the Kodak rows in Table II are partly training-set performance. Additionally, Sec. IV-A substitutes no-reference perceptual metrics for PSNR/SSIM because erasure degrades reference-based metrics, which makes the headline quality comparison depend on metrics that may reward smoothing rather than fidelity. The self-citations to the authors' prior lossless-compression and FPGA works are not load-bearing for the central compression claim. Overall, the paper's efficiency and flexibility contributions have independent content, but a substantial part of the claimed quality improvement is confounded by test-set fine-tuning, giving a partial circularity score of 6.

Assumptions & free parameters 4 free parameters · 6 assumptions · 0 invented entities

The central claims rest mainly on the ability of a small transformer to reconstruct erased patches and on using no-reference metrics as the quality yardstick. Free parameters include the loss weight, pretraining erase ratio, patch dimensions, and row-sampler thresholds (the latter not reported). No new physical entities are introduced. The main independent-support concern is test-set fine-tuning: the model is fine-tuned on Kodak in Sec IV-D and then scored on Kodak in Table II.

free parameters (4)

Loss weight lambda in Eq (2) = 0.3
Manually chosen to balance L1 and LPIPS losses; no sensitivity analysis is reported, so all quality numbers are contingent on this choice.
Erase ratio during pretraining = 0.25
Offline pretraining uses an erase ratio of 0.25; reported reconstruction quality may depend on this value.
Patch sizes n and b = n=32, b=1/2/4 in experiments
The two-stage patchify dimensions control the complexity-quality tradeoff; the ablation recommends a particular patch size, so results are tuned to this choice.
Row sampler thresholds delta and Delta = not reported
The constraints in the row-based sampler (Eq 1) are central to mask generation, but no numeric values are given, making replication impossible and the effect on quality unquantified.

assumptions (6)

domain assumption Erased image content can be recovered accurately by the transformer from the squeezed context.
Underlies the quality claim; no reference-metric validation is provided because erasure lowers PSNR and SSIM (Sec IV-A).
domain assumption No-reference metrics BRISQUE, PI, and TRES are reliable indicators of compression quality at matched BPP.
Used for all codec comparisons; the paper does not show correlation with human ratings or reference metrics.
domain assumption CIFAR-10 pretraining plus Kodak fine-tuning transfers to test images.
Training procedure in Sec IV-A and fine-tuning in Sec IV-D; Kodak is then used as a test set, which violates independence for Kodak results.
domain assumption Two-stage patchify with local attention preserves enough global context for reconstruction.
The complexity reduction in Sec III-B restricts attention to n x n patches, assuming local context suffices.
domain assumption The binary erase mask can be transmitted cheaply and exactly.
The paper states a 32x32 mask is 128 bytes; reconstruction relies on mask positions being known at the receiver.
standard math Transformer self-attention cost scales quadratically with the number of tokens.
Relied on in the complexity analysis of Sec III-B to justify the two-stage patchify design.

how reviews work

0 comments

Cite this review

Pith. "Pith review of Easz: An Agile Transformer-based Image Compression Framework for Resource-constrained IoTs." pith.science (2026). https://pith.science/paper/FVUOWU6L

@misc{pith2026250501742,
  author       = {Pith},
  title        = {Pith review of: Easz: An Agile Transformer-based Image Compression Framework for Resource-constrained IoTs},
  year         = {2026},
  howpublished = {\url{https://pith.science/paper/FVUOWU6L}},
  note         = {Machine review of arXiv:2505.01742}
}

read the original abstract

Neural image compression, necessary in various machine-to-machine communication scenarios, suffers from its heavy encode-decode structures and inflexibility in switching between different compression levels. Consequently, it raises significant challenges in applying the neural image compression to edge devices that are developed for powerful servers with high computational and storage capacities. We take a step to solve the challenges by proposing a new transformer-based edge-compute-free image coding framework called Easz. Easz shifts the computational overhead to the server, and hence avoids the heavy encoding and model switching overhead on the edge. Easz utilizes a patch-erase algorithm to selectively remove image contents using a conditional uniform-based sampler. The erased pixels are reconstructed on the receiver side through a transformer-based framework. To further reduce the computational overhead on the receiver, we then introduce a lightweight transformer-based reconstruction structure to reduce the reconstruction load on the receiver side. Extensive evaluations conducted on a real-world testbed demonstrate multiple advantages of Easz over existing compression approaches, in terms of adaptability to different compression levels, computational efficiency, and image reconstruction quality.

Figures

Figures reproduced from arXiv: 2505.01742 by the authors.

**Figure 4.** Easz preserves details better than SR methods via direct [PITH_FULL_IMAGE:figures/full_fig_p003_4.png] view at source ↗

**Figure 5.** Reconstruct process illustration. remove k random sub-patches. The encoder E receives the remaining un-erased sub-patches U = {u1, u2, . . . , um} where m = n/b × n/b − k. Each un-erased sub-patch ui is projected into an embedding ei . The two-layer encoder processes these embeddings to extract features: F = E(e1, e2, . . . , em). To compensate for erased blocks during reconstruction, we introduce zero vectors Fˆ =… view at source ↗

Figures from the paper (2 more)

**Figure 6.** Figure 6: Efficiency Evaluation on NVIDIA Jetson TX2. [PITH_FULL_IMAGE:figures/full_fig_p005_6.png]

**Figure 7.** Figure 7: (a)(b): Comparison between Easz with proposed mask [PITH_FULL_IMAGE:figures/full_fig_p006_7.png]

Discussion (0). Continue with ORCID to comment.

Reference graph

Works this paper leans on

34 extracted references · 28 canonical work pages

[1]

Towards drone-sourced live video analytics for the construction in- dustry,

S. George, J. Wang, M. Bala, T. Eiszler, P. Pillai, and M. Satyanarayanan, “Towards drone-sourced live video analytics for the construction in- dustry,” in Proceedings of the 20th International Workshop on Mobile Computing Systems and Applications , pp. 3–8, ACM, 2019

work page 2019
[2]

Bringing Cutting-Edge Technology to Wildlife Conservation

“Bringing Cutting-Edge Technology to Wildlife Conservation.” https: //www.wildlifeinsights.org/, 2023

work page 2023
[3]

Research on the application of computer vision based on deep learning in autonomous driving technology,

J. Zhang, J. Cao, J. Chang, X. Li, H. Liu, and Z. Li, “Research on the application of computer vision based on deep learning in autonomous driving technology,” in International conference on wireless communi- cations networking and applications , pp. 82–91, Springer, 2023

work page 2023
[4]

Compressive sensing based asymmetric semantic image compression for resource- constrained iot system,

Y . Huang, B. Chen, J. Zhang, Q. Han, and S.-T. Xia, “Compressive sensing based asymmetric semantic image compression for resource- constrained iot system,” in Proceedings of the 59th ACM/IEEE Design Automation Conference, pp. 877–882, 2022

work page 2022
[5]

Group, “JPEG.” https://jpeg.org/, 1986

J. Group, “JPEG.” https://jpeg.org/, 1986

work page 1986
[6]

Bellard, “BPG.” https://bellard.org/bpg/, 2014

F. Bellard, “BPG.” https://bellard.org/bpg/, 2014

work page 2014
[7]

The future of computing is distributed,

“The future of computing is distributed,” 2020. [Accessed on October 14, 2023]

work page 2020
[8]

Ai and compute,

“Ai and compute,” 2018

work page 2018

Show all 34 references

[9]

Moby: Empowering 2D Models for Efficient Point Cloud Analytics on the Edge,

J. Li, Y . H. Cai, L. Liu, Y . Mao, C. J. Xue, and H. Xu, “Moby: Empowering 2D Models for Efficient Point Cloud Analytics on the Edge,” in Proc. ACM MM , 2023

2023
[10]

Cross-Camera Inference on the Constrained Edge,

J. Li, L. Liu, H. Xu, S. Wu, and C. J. Xue, “Cross-Camera Inference on the Constrained Edge,” in Proc. IEEE INFOCOM , 2023

2023
[11]

Gaussian rate-distortion- perception coding and entropy-constrained scalar quantization,

L. Xie, L. Li, J. Chen, L. Yu, and Z. Zhang, “Gaussian rate-distortion- perception coding and entropy-constrained scalar quantization,” arXiv preprint arXiv:2409.02388, 2024

2024 arXiv
[12]

Output-constrained lossy source coding with application to rate-distortion-perception theory,

L. Xie, L. Li, J. Chen, and Z. Zhang, “Output-constrained lossy source coding with application to rate-distortion-perception theory,” IEEE Transactions on Communications , 2024

2024
[13]

Joint autoregressive and hier- archical priors for learned image compression,

D. Minnen, J. Ball ´e, and G. Toderici, “Joint autoregressive and hier- archical priors for learned image compression,” in NeurIPS, Montr ´eal, Canada, pp. 10794–10803, 2018

2018
[14]

Learned image com- pression with discretized gaussian mixture likelihoods and attention modules,

Z. Cheng, H. Sun, M. Takeuchi, and J. Katto, “Learned image com- pression with discretized gaussian mixture likelihoods and attention modules,” in CVPR, 2020

2020
[15]

Faster and stronger lossless compression with optimized autoregressive framework,

Y . Mao, J. Li, Y . Cui, and J. C. Xue, “Faster and stronger lossless compression with optimized autoregressive framework,” in 2023 60th ACM/IEEE Design Automation Conference (DAC) , pp. 1–6, 2023

2023
[16]

Accelerating general-purpose lossless compression via simple and scalable parameterization,

Y . Mao, Y . Cui, T.-W. Kuo, and C. J. Xue, “Accelerating general-purpose lossless compression via simple and scalable parameterization,” in Proceedings of the 30th ACM International Conference on Multimedia , pp. 3205–3213, 2022

2022
[17]

Trace: A fast transformer- based general-purpose lossless compressor,

Y . Mao, Y . Cui, T.-W. Kuo, and C. J. Xue, “Trace: A fast transformer- based general-purpose lossless compressor,” in Proceedings of the ACM Web Conference 2022 , pp. 1829–1838, 2022

2022
[18]

Ariadne: A hotness-aware and size-adaptive compressed swap tech- nique for fast application relaunch and reduced cpu usage on mobile devices,

Y . Liang, A. Shen, C. J. Xue, R. Pan, H. Mao, N. M. Ghiasi, Q. Jiang, R. Nadig, L. Li, R. Ausavarungnirun, M. Sadrosadati, and O. Mutlu, “Ariadne: A hotness-aware and size-adaptive compressed swap tech- nique for fast application relaunch and reduced cpu usage on mobile devic...

2025
[19]

Hirise: High-resolution image scaling for edge ml via in- sensor compression and selective roi,

B. Reidy, S. Tabrizchi, M. Mohammadi, S. Angizi, A. Roohi, and R. Zand, “Hirise: High-resolution image scaling for edge ml via in- sensor compression and selective roi,” in Proceedings of the 61st ACM/IEEE Design Automation Conference , pp. 1–6, 2024

2024
[20]

Online streaming video super-resolution with convolutional look-up table,

G. Yin, Z. Qu, X. Jiang, S. Jiang, Z. Han, N. Zheng, X. Liu, H. Yang, Y . Yang, D. Li, et al. , “Online streaming video super-resolution with convolutional look-up table,” arXiv preprint arXiv:2303.00334 , 2023

2023 arXiv
[21]

Deep model-based super- resolution with non-uniform blur,

C. Laroche, A. Almansa, and M. Tassano, “Deep model-based super- resolution with non-uniform blur,” in Proceedings of the IEEE/CVF winter conference on applications of computer vision , pp. 1797–1808, 2023

2023
[22]

Mapping new realities: Ground truth image creation with pix2pix image-to-image translation,

Z. Li, B. Guan, Y . Wei, Y . Zhou, J. Zhang, and J. Xu, “Mapping new realities: Ground truth image creation with pix2pix image-to-image translation,” arXiv preprint arXiv:2404.19265 , 2024

2024 arXiv
[23]

On the compressibility of quantized large language models,

Y . Mao, W. Wang, H. Du, N. Guan, and C. J. Xue, “On the compressibility of quantized large language models,” arXiv preprint arXiv:2403.01384, 2024

2024 arXiv
[24]

Attention is all you need,

A. Vaswani, N. Shazeer, N. Parmar, J. Uszkoreit, L. Jones, A. N. Gomez, Ł. Kaiser, and I. Polosukhin, “Attention is all you need,” Advances in neural information processing systems , vol. 30, 2017

2017
[25]

The unreasonable effectiveness of deep features as a perceptual metric,

R. Zhang, P. Isola, A. A. Efros, E. Shechtman, and O. Wang, “The unreasonable effectiveness of deep features as a perceptual metric,” in Proceedings of the IEEE conference on computer vision and pattern recognition, pp. 586–595, 2018

2018
[26]

Learning multiple layers of features from tiny images,

K. Alex and H. Geoffrey, “Learning multiple layers of features from tiny images,” 2009, 2009

2009
[27]

Kodak Lossless True Color Image Suite

E. K. Company, “Kodak Lossless True Color Image Suite.” https://r0k. us/graphics/kodak/, 1993

1993
[28]

Challenge on Learned Image Compression

Workshop and C. on Learned Image Compression, “Challenge on Learned Image Compression.” https://compression.cc/, 2022

2022
[29]

No-reference image quality assessment in the spatial domain,

A. Mittal, A. K. Moorthy, and A. C. Bovik, “No-reference image quality assessment in the spatial domain,” IEEE Transactions on image processing, vol. 21, no. 12, pp. 4695–4708, 2012

2012
[30]

The 2018 pirm challenge on perceptual image super-resolution,

Y . Blau, R. Mechrez, R. Timofte, T. Michaeli, and L. Zelnik-Manor, “The 2018 pirm challenge on perceptual image super-resolution,” in ECCV Workshops, 2018

2018
[31]

No-reference image quality assessment via transformers, relative ranking and self- consistency,

S. A. Golestaneh, S. Dadsetan, and K. M. Kitani, “No-reference image quality assessment via transformers, relative ranking and self- consistency,” in Proceedings of the IEEE/CVF Winter Conference on Applications of Computer Vision , pp. 1220–1230, 2022

2022
[32]

p lpaq: Acceler- ating lpaq compression on fpga,

D. Tang, X. Sun, N. Guan, T.-W. Kuo, and C. J. Xue, “ p lpaq: Acceler- ating lpaq compression on fpga,” in 2022 International Conference on Field-Programmable Technology (ICFPT), pp. 1–6, IEEE, 2022

2022
[33]

Stem: Streaming-based fpga acceleration for large-scale compactions in lsm kv,

D. Tang, W. Wang, Y . Mao, J. Yu, T.-W. Kuo, and C. J. Xue, “Stem: Streaming-based fpga acceleration for large-scale compactions in lsm kv,” in 2024 IEEE 40th International Conference on Data Engineering (ICDE), pp. 3893–3905, IEEE, 2024

2024
[34]

Tegrastats Utility

“Tegrastats Utility.” https://docs.nvidia.com/drive/drive os 5.1.6.1L/ nvvib docs/DRIVE OS Linux SDK Development Guide/Utilities/ util tegrastats.html, 2023

2023

Pith tools

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

[1] [1]

Towards drone-sourced live video analytics for the construction in- dustry,

S. George, J. Wang, M. Bala, T. Eiszler, P. Pillai, and M. Satyanarayanan, “Towards drone-sourced live video analytics for the construction in- dustry,” in Proceedings of the 20th International Workshop on Mobile Computing Systems and Applications , pp. 3–8, ACM, 2019

work page 2019

[2] [2]

Bringing Cutting-Edge Technology to Wildlife Conservation

“Bringing Cutting-Edge Technology to Wildlife Conservation.” https: //www.wildlifeinsights.org/, 2023

work page 2023

[3] [3]

Research on the application of computer vision based on deep learning in autonomous driving technology,

J. Zhang, J. Cao, J. Chang, X. Li, H. Liu, and Z. Li, “Research on the application of computer vision based on deep learning in autonomous driving technology,” in International conference on wireless communi- cations networking and applications , pp. 82–91, Springer, 2023

work page 2023

[4] [4]

Compressive sensing based asymmetric semantic image compression for resource- constrained iot system,

Y . Huang, B. Chen, J. Zhang, Q. Han, and S.-T. Xia, “Compressive sensing based asymmetric semantic image compression for resource- constrained iot system,” in Proceedings of the 59th ACM/IEEE Design Automation Conference, pp. 877–882, 2022

work page 2022

[5] [5]

Group, “JPEG.” https://jpeg.org/, 1986

J. Group, “JPEG.” https://jpeg.org/, 1986

work page 1986

[6] [6]

Bellard, “BPG.” https://bellard.org/bpg/, 2014

F. Bellard, “BPG.” https://bellard.org/bpg/, 2014

work page 2014

[7] [7]

The future of computing is distributed,

“The future of computing is distributed,” 2020. [Accessed on October 14, 2023]

work page 2020

[8] [8]

Ai and compute,

“Ai and compute,” 2018

work page 2018