Pith. sign in

REVIEW 5 major objections 4 minor 48 references

Unified Coding for Both Human Perception and Generalized Machine Analytics with CLIP Supervision

T0 review · 5 major / 4 minor · reviewed 2026-08-10 · deepseek-v4-flash

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

Pith's one-line read This paper proposes a unified image codec that serves both human perception and unseen machine analytics from a single bitstream, trained without task-specific supervision.

desk verdict A plausible and genuinely novel CLIP-based unified image coding scheme whose reported gains are real but not yet statistically pinned down; deserves a proper referee, not a desk reject. read the letter →

arxiv 2501.04579 v1 pith:SAG5HWWN submitted 2025-01-08 cs.CV cs.MM

Kangsheng Yin , Quan Liu , Xuelin Shen , Yulin He , Wenhan Yang , Shiqi Wang This is my paper · ORCID
classification cs.CVcs.MM
keywords imagecompressioncodingformachineCLIPsupervisionlearnedconditionaldecodingmulti-scalesemanticlosshuman-machineunifiedgeneralization
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 proposes UG-ICM, an image compression scheme that trains one codec to serve two audiences at once: human viewers and machine analytics models, including models not seen during training. The key move is to use a pre-trained contrastive vision-language embedding as the semantic referee, pushing compressed images to stay close to the originals at whole-image, local-patch, and instance levels. A preference-conditioned decoder then turns one shared bitstream into either a human-pleasing reconstruction or a machine-oriented one, selected by a single tunable parameter. On the MLIC++ backbone at matched bits-per-pixel, the machine-oriented version improves classification, object detection, semantic segmentation, and instance segmentation by average bpp-analytics margins of 2.8%, 3.9%, 6.0%, and 3.2%, while the human-oriented version keeps PSNR, SSIM, and LPIPS on par with the backbone. If correct, this means task-agnostic semantic coding is achievable without task-specific labels or multiple bitstreams.

What carries the argument

The load-bearing machinery is the Multi-Scale CLIP loss and the Preference Conditional Decoding Module (PCDM). The MS-CLIP loss measures cosine similarity between the original and reconstructed images in the CLIP image-embedding space at three granularities: the full image (global), random local crops (local), and instance crops produced by a segmentation model (instance); summing the three pushes the codec to preserve hierarchical semantics. The PCDM is a lightweight module placed in each decoding block: a two-layer MLP turns the preference indicator beta in {beta_h, beta_m} into a bias feature that is added to the decoder's image features, steering reconstruction toward human or machine utility. A two-stage training schedule alternates human-oriented and machine-oriented objectives so the encoder produces a unified latent and the decoder learns the two branches. Together they make a single bitstream that can be decoded in two different versions without recompression.

What would settle it

Pick an analytics task whose success is known to be poorly captured by CLIP-style global semantics, for instance monocular depth estimation or counting of small objects, and run the machine-preference decode of UG-ICM on it at matched bitrates. If the bpp-analytics gains over the vanilla codec disappear or reverse on that task, the claimed task-agnostic generalization is bounded by CLIP's semantic coverage; if the gains persist, the generalized claim is stronger than the paper's current evaluation shows.

Watch

Extended reading notes

Core claim

The paper claims that a learned image codec can be made simultaneously human- and machine-friendly, and generalizable to unseen analytics, by two mechanisms. First, it replaces task-specific supervision with multi-scale CLIP-based semantic supervision: a cosine-similarity loss at global, local-crop, and instance-crop levels between the original and reconstructed images, computed in the embedding space of a pre-trained CLIP image encoder. Second, a Preference Conditional Decoding Module injects a preference feature into each decoder block, so the same latent bitstream produces a human-preferred image when the preference parameter is beta_h and a machine-preferred image when it is beta_m. The whole model is trained self-supervised and alternates between rate-perception and rate-analytics objectives; at test time the downstream analytics models are unseen. On the MLIC++ backbone the machine decode gives average bpp-based gains of 2.8% for classification, 3.9% for object detection, 6.0% for semantic segmentation, and 3.2% for instance segmentation over the vanilla codec, with comparable PSNR, SSIM, and LPIPS.

Load-bearing premise

The method assumes that closeness in the CLIP image-embedding space, at global, local, and instance scales, is a faithful proxy for what later, unseen machine analytics tasks need; if that proxy fails for a task, the training signal will not help that task.

Editorial extensions

If this is right

  • A single compressed bitstream can be decoded into a human version or a machine version by switching one scalar preference, so separate human and machine streams become unnecessary.
  • Because training needs no downstream labels or analytics models, the same recipe can be re-run when new tasks arrive, without retraining the codec for each task.
  • The reported gains on unseen classification, detection, and segmentation suggest that semantic preservation in a generic embedding space transfers across task families, not just to tasks used for evaluation.
  • The multi-scale variant outperforms global-only CLIP supervision, indicating that small-object and local semantics are material to machine analytics and are worth explicit loss terms.
  • The recipe transfers across two learned codec backbones, so it acts as a training-time add-on rather than a redesign of the entropy model or architecture.

Reading between the lines

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

  • Implicit extension: because CLIP embeddings are aligned with text, a language-conditioned variant could steer the machine decode toward a specified task family, such as pedestrians or road markings, without retraining the codec; the paper does not explore this.
  • Testable extension: the paper's ablations isolate the joint effect of the three CLIP losses but not their optimal balance; replacing the fixed equal weighting with learned or rate-dependent weights could yield larger gains.
  • Boundary inference: the mechanism should help any analytics task whose errors align with CLIP's notion of semantic similarity, such as common recognition tasks, but tasks like geometric measurement or fine-grained counting may see little or no benefit; the paper's task set does not cover those cases.
  • Consequence for codec design: if the result holds, learned codecs for machine consumption can drop task-specific heads and instead budget bits toward preserving human-interpretable semantics, making the bitstream a shared resource rather than a task-specific artifact.
Share X Bluesky LinkedIn Reddit HN

Editorial analysis

A structured set of objections, weighed in public.

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

Referee Report

5 major / 4 minor

Summary. The paper proposes UG-ICM, a learned image compression framework that combines a preference-conditional decoder with CLIP-based multi-scale semantic supervision. The encoder and entropy model produce a unified bitstream, and the decoder takes a scalar preference code (human or machine) to produce two reconstructions. The training alternates between rate-perception and rate-analytics objectives, and the CLIP losses are applied at global, random-local, and instance levels (Eqs. 4-7), with instance masks obtained from FreeSOLO. Experiments on two LIC backbones, Bmshj2018 and MLIC++, report bpp-analytics gains of 1.7-6.0% on four downstream tasks, comparable human-perceptual quality, and ablations of the PCDM and multi-scale loss.

Significance. The manuscript addresses a relevant and active topic, and the central idea is plausible: using a frozen multimodal model as task-agnostic supervision and conditioning the decoder on a preference code could provide a practical single-bitstream solution for both human and machine consumption. The paper provides a code link, evaluates on four downstream tasks, and uses two different backbones, which is a reasonable breadth of evidence for a conference submission. The main value is empirical; if the gains are robust, the method would be a useful contribution to the image-coding-for-machines literature. The claims of 'remarkable' and 'unseen' generalization are, however, currently stronger than the evidence supports.

major comments (5)
  1. [Experimental Results, Figs. 5 and 8] The reported average bpp-analytics gains are not accompanied by any measure of variance, and the phrase 'average bpp-analytics gain' is never defined. The Implementation Details state that models are trained four times with different Lagrange parameters; that is four rate points, not four random seeds. Because learned compression training is sensitive to initialization and the claimed improvements are 1.7-6.0%, the central claim that CLIP supervision improves unseen analytics is not statistically established. Please provide multiple seeds and error bars, and specify how equal-bpp comparisons and the reported averages are computed, including the interpolation rule and bpp anchors.
  2. [Proposed Method, Eq. (4)] The instance-level CLIP loss pairs a set of masks from the original image with a set of masks from the compressed image, but the paper does not explain how the two sets are matched. FreeSOLO can return different numbers or arrangements of masks for the original and the compressed image; if pairing is only by index, the loss may align unrelated instances. Please describe the mask extraction and matching procedure, including confidence thresholds, maximum number of masks, and whether masks are taken from the machine-preference reconstruction at every training step, or compute the loss using masks transferred from the original image only.
  3. [Proposed Method, Eq. (5) and Implementation Details] The local-wise CLIP loss is underspecified: the paper does not state the crop size, the number of local crops per image, whether the same spatial coordinates are used for the original and reconstructed crops, or how crops are resized for the CLIP ViT-B/32 input. These choices directly affect the learned representation and the reported gains, so they should be documented to make the MS-CLIP loss reproducible.
  4. [Comparison to State-of-the-art ICM Methods, Fig. 8] The paper notes that TransTIC requires recompression into another bitstream when adapting to a new task, but it does not state whether the bpp plotted for TransTIC in Fig. 8 includes both the human-oriented and the analytics-oriented bitstreams. If only one bitstream is counted, the comparison is biased in favor of UG-ICM, whose advantage is precisely the use of a single bitstream. Please clarify the bitrate accounting for all anchor methods.
  5. [Abstract and Benchmark] The statement that instance segmentation is an 'unseen' task is imprecise: the MS-CLIP loss uses FreeSOLO, which is itself an instance segmentation model, to generate the instance-level pairs. The compression model therefore receives instance-level segmentation information during training, even though the evaluation model, Mask R-CNN, and its annotations are not used. Please either weaken the 'unseen task' claim or explain why FreeSOLO's outputs should not be counted as task supervision.
minor comments (4)
  1. [Throughout] Please fix typos such as 'pretraind', 'tranining', 'Multy-Scale', and 'glocal-local-instance'.
  2. [Eq. (3)] The cosine term has a mismatched bracket; please correct the notation.
  3. [Fig. 9] The text 'average gain of3%' and '1dB bpp-PSNR' should be formatted as '3%' and '1 dB', and the terms bpp-mAP and bpp-PSNR should be defined.
  4. [Implementation Details] Please report the batch size, training image resolution, number of local crops per image, and the preprocessing applied before CLIP feature extraction.

Circularity Check

0 steps flagged · score 0.0 of 10

No significant circularity: CLIP supervision is an external training signal and all reported gains are measured on independent downstream models.

full rationale

The paper's claimed derivation is not circular. The training objective, MS-CLIP loss, is defined as cosine similarity between CLIP features of original and reconstructed images at global, local, and instance levels (Eqs. 4-7), while the reported evaluation metrics are Top-1 accuracy, mAP, mIoU, PSNR, SSIM, and LPIPS computed by external task models (ResNet101, YOLOv3, Mask R-CNN, DeepLabv3+) and standard image quality measures. None of these evaluation quantities appears in the training loss, and the downstream models are not used to construct the training objective, so the reported gains are not forced by construction. The paper does not fit a parameter and then rename it as a prediction; it trains a compression model and measures generalization on held-out tasks. The one overlapping-author citation, Shen et al. 2024, appears only in a related-work survey sentence ('transfer-based methods ... (Shen et al. 2024)') and is not load-bearing for the proposed method or its evaluation. No uniqueness theorem is imported from the authors' prior work, and no ansatz is smuggled in via citation. The absence of error bars or significance tests is a legitimate robustness concern, but it is not a circularity: the empirical comparison is externally grounded. The use of FreeSOLO to generate instance masks for the CLIP instance-level loss does overlap topically with the instance-segmentation evaluation, but the evaluation uses a different, externally trained model (Mask R-CNN) and reports its mAP, not a CLIP similarity score; therefore the reported instance-segmentation improvement is not the training loss by another name. Overall, the derivation chain is self-contained and externally benchmarked, so the circularity score is 0.

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

The central claim rests on the transferability of CLIP features and the reliability of the auxiliary segmentation model. No new physical or mathematical entities are introduced. The free parameters are standard hyperparameters and a few design choices that are not rigorously tuned.

free parameters (4)
  • Lagrange parameter lambda = four values (not explicitly listed)
    Used in the rate-distortion objective to obtain different bitrate operating points. The choice of four values is empirical.
  • Preference codes beta_h and beta_m = [0, 1]
    These are chosen as 0 and 1 to represent human and machine preferences, a design decision.
  • Loss weighting for global, local, instance CLIP losses = equal weights (implied by Eq. 7)
    The total MS-CLIP loss simply sums the three components, with no learned or tuned weights, which is a simplifying assumption.
  • Training epochs and learning rates = 200 epochs at 1e-4, then 10 epochs at 1e-5
    These hyperparameters are chosen by hand and are not derived from any analysis.
assumptions (4)
  • domain assumption CLIP feature space is a transferable semantic representation for a wide range of unseen machine analytics tasks.
    The entire generalization claim relies on CLIP embeddings capturing task-generic semantics. This is a widely known empirical property, but the paper does not prove or bound it.
  • domain assumption The instance segmentation model FreeSOLO provides consistent and reliable instance masks for both original and compressed images.
    Eq. 4 assumes a set of instance-wise pairs can be obtained; if the segmentation on the compressed image changes drastically, the loss may misalign. The paper does not analyze this risk.
  • domain assumption L2 reconstruction loss is an adequate proxy for human perceptual quality during training.
    The human preference training uses MSE loss, while evaluation uses PSNR, SSIM, and LPIPS. The paper does not justify why MSE is sufficient for human perception.
  • standard math Lagrangian relaxation for rate-distortion optimization is a standard and valid training objective.
    The training losses follow the conventional rate-distortion Lagrangian formulation used throughout learned image compression.

how reviews work

0 comments
Cite this review

Pith. "Pith review of Unified Coding for Both Human Perception and Generalized Machine Analytics with CLIP Supervision." pith.science (2026). https://pith.science/paper/SAG5HWWN

@misc{pith2026250104579,
  author       = {Pith},
  title        = {Pith review of: Unified Coding for Both Human Perception and Generalized Machine Analytics with CLIP Supervision},
  year         = {2026},
  howpublished = {\url{https://pith.science/paper/SAG5HWWN}},
  note         = {Machine review of arXiv:2501.04579}
}
read the original abstract

The image compression model has long struggled with adaptability and generalization, as the decoded bitstream typically serves only human or machine needs and fails to preserve information for unseen visual tasks. Therefore, this paper innovatively introduces supervision obtained from multimodal pre-training models and incorporates adaptive multi-objective optimization tailored to support both human visual perception and machine vision simultaneously with a single bitstream, denoted as Unified and Generalized Image Coding for Machine (UG-ICM). Specifically, to get rid of the reliance between compression models with downstream task supervision, we introduce Contrastive Language-Image Pre-training (CLIP) models into the training constraint for improved generalization. Global-to-instance-wise CLIP supervision is applied to help obtain hierarchical semantics that make models more generalizable for the tasks relying on the information of different granularity. Furthermore, for supporting both human and machine visions with only a unifying bitstream, we incorporate a conditional decoding strategy that takes as conditions human or machine preferences, enabling the bitstream to be decoded into different versions for corresponding preferences. As such, our proposed UG-ICM is fully trained in a self-supervised manner, i.e., without awareness of any specific downstream models and tasks. The extensive experiments have shown that the proposed UG-ICM is capable of achieving remarkable improvements in various unseen machine analytics tasks, while simultaneously providing perceptually satisfying images.

Figures

Figures reproduced from arXiv: 2501.04579 by the authors.

Figure 1
Figure 1. (a) Most existing ICM models rely on task-specific [PITH_FULL_IMAGE:figures/full_fig_p001_1.png] view at source ↗
Figure 2
Figure 2. (a) Compressing pipeline of the proposed UG-ICM. (b) Details of the proposed PCDM. (c) Modules and loss terms [PITH_FULL_IMAGE:figures/full_fig_p003_2.png] view at source ↗
Figure 3
Figure 3. Empirical analysis on generalization of taking [PITH_FULL_IMAGE:figures/full_fig_p003_3.png] view at source ↗
Figures from the paper (7 more)
Figure 4
Figure 4. Figure 4: Illustration of the proposed Multi-Scale CLIP loss. [PITH_FULL_IMAGE:figures/full_fig_p004_4.png]
Figure 5
Figure 5. Figure 5: Machine analytics performance comparisons between the proposed UG-ICM and the backbone compression networks. [PITH_FULL_IMAGE:figures/full_fig_p005_5.png]
Figure 6
Figure 6. Figure 6: Perceptual quality comparisons between the proposed UG-ICM and the backbone compression networks. [PITH_FULL_IMAGE:figures/full_fig_p005_6.png]
Figure 7
Figure 7. Figure 7: Visualization of the machine analytics performance regarding classification (a), object detection (b), and instance [PITH_FULL_IMAGE:figures/full_fig_p006_7.png]
Figure 5
Figure 5. Figure 5: Encouraging observations have been made, showing [PITH_FULL_IMAGE:figures/full_fig_p006_5.png]
Figure 8
Figure 8. Figure 8: Performance comparison between the proposed UG-ICM and employed anchors regarding classification (a), semantic [PITH_FULL_IMAGE:figures/full_fig_p007_8.png]
Figure 9
Figure 9. Figure 9: Performance comparisons between decoded im [PITH_FULL_IMAGE:figures/full_fig_p007_9.png]

Discussion (0). Continue with ORCID to comment.

Reference graph

Works this paper leans on

48 extracted references · 42 canonical work pages

  1. [1]

    Ball \'e , J.; Laparra, V.; and Simoncelli, E. P. 2017. End-to-end optimized image compression. In Proceedings of the International Conference on Learning Representations

    work page 2017

  2. [2]

    J.; and Johnston, N

    Ball \'e , J.; Minnen, D.; Singh, S.; Hwang, S. J.; and Johnston, N. 2018. Variational image compression with a scale hyperprior. In Proceedings of the International Conference on Learning Representations

    work page 2018

  3. [3]

    Bross, B.; Wang, Y.; Ye, Y.; Liu, S.; Chen, J.; Sullivan, G.; and Ohm, J. 2021. Overview of the Versatile Video Coding ( VVC ) Standard and its Applications. IEEE Transactions on Circuits and Systems for Video Technology, 31(10): 3736--3764

    work page 2021

  4. [4]

    D.; Racap \'e , F.; B \'e gaint, J.; Pushparaja, A.; and Feltman, S

    Chamain, L. D.; Racap \'e , F.; B \'e gaint, J.; Pushparaja, A.; and Feltman, S. 2021. End-to-end optimized image compression for machines, a study. In Proceedings of the Data Compression Conference, 163--172

    work page 2021

  5. [5]

    Chen, F.; Xu, Y.; and Wang, L. 2022. Two-stage octave residual network for end-to-end image compression. In Proceedings of the AAAI Conference on Artificial Intelligence, volume 36, 3922--3929

    work page 2022

  6. [6]

    Chen, L.-C.; Zhu, Y.; Papandreou, G.; Schroff, F.; and Adam, H. 2018. Encoder-decoder with atrous separable convolution for semantic image segmentation. In Proceedings of the European Conference on Computer Vision, 801--818

    2018

  7. [7]

    Chen, Y.-H.; Weng, Y.-C.; Kao, C.-H.; Chien, C.; Chiu, W.-C.; and Peng, W.-H. 2023. Transtic: Transferring transformer-based image compression from human perception to machine perception. In Proceedings of the IEEE International Conference on Computer Vision, 23297--23307

    work page 2023

  8. [8]

    Chen, Z.; Fan, K.; Wang, S.; Duan, L.; Lin, W.; and Kot, A. 2020. Toward Intelligent Sensing: Intermediate Deep Feature Compression. IEEE Transactions on Image Processing, 29: 2230--2243

    work page 2020

Show all 48 references
  1. [9]

    Chen, Z.; Fan, K.; Wang, S.; Duan, Y.; Lin, W.; and Alex, K. 2019. Lossy intermediate deep learning feature compression and evaluation. In Proceedings of the ACM International Conference on Multimedia, 2414--2422

    2019

  2. [10]

    Choi, H.; and Baji \'c , I. V. 2022. Scalable image coding for humans and machines. IEEE Transactions on Image Processing, 31: 2739--2754

    2022

  3. [11]

    Choi, J.; and Han, B. 2020. Task-aware quantization network for jpeg image compression. In Proceedings of European Conference on Computer Vision, 309--324

    2020

  4. [12]

    Deng, J.; Dong, W.; Socher, R.; Li, L.-J.; Li, K.; and Fei-Fei, L. 2009. Imagenet: A large-scale hierarchical image database. In Proceedings of the IEEE Conference on Computer Vision and Pattern Recognition, 248--255

    2009

  5. [13]

    K.; Winn, J.; and Zisserman, A

    Everingham, M.; Van Gool, L.; Williams, C. K.; Winn, J.; and Zisserman, A. 2010. The pascal visual object classes (voc) challenge. International Journal of Computer Vision, 88: 303--338

    2010

  6. [14]

    Gao, Y.; Liu, J.; Xu, Z.; Zhang, J.; Li, K.; Ji, R.; and Shen, C. 2022. Pyramidclip: Hierarchical feature alignment for vision-language model pretraining. Neural Information Processing Systems, 35: 35959--35970

    2022

  7. [15]

    Guo, Z.; Zhang, Z.; Feng, R.; and Chen, Z. 2021. Causal contextual prediction for learned image compression. IEEE Transactions on Circuits and Systems for Video Technology, 32(4): 2329--2341

    2021

  8. [16]

    He, D.; Zheng, Y.; Sun, B.; Wang, Y.; and Qin, H. 2021. Checkerboard context model for efficient learned image compression. In Proceedings of the IEEE Conference on Computer Vision and Pattern Recognition, 14771--14780

    2021

  9. [17]

    He, K.; Gkioxari, G.; Doll \'a r, P.; and Girshick, R. 2017. Mask r-cnn. In Proceedings of the IEEE International Conference on Computer Vision, 2961--2969

    2017

  10. [18]

    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

    2016

  11. [19]

    Huang, Z.; Jia, C.; Wang, S.; and Ma, S. 2021. Visual analysis motivated rate-distortion model for image coding. In Proceedings of IEEE International Conference on Multimedia and Expo, 1--6

    2021

  12. [20]

    Jiang, W.; and Wang, R. 2023. MLIC++: Linear Complexity Multi-Reference Entropy Modeling for Learned Image Compression. In ICML Workshop Neural Compression: From Information Theory to Applications

    2023

  13. [21]

    Jiang, W.; Yang, J.; Zhai, Y.; Ning, P.; Gao, F.; and Wang, R. 2023. Mlic: Multi-reference entropy model for learned image compression. In Proceedings of the ACM International Conference on Multimedia, 7618--7627

    2023

  14. [22]

    C.; Lo, W.-Y.; et al

    Kirillov, A.; Mintun, E.; Ravi, N.; Mao, H.; Rolland, C.; Gustafson, L.; Xiao, T.; Whitehead, S.; Berg, A. C.; Lo, W.-Y.; et al. 2023. Segment anything. In Proceedings of the IEEE/CVF International Conference on Computer Vision, 4015--4026

    2023

  15. [23]

    Le, N.; Zhang, H.; Cricri, F.; Ghaznavi-Youvalari, R.; and Rahtu, E. 2021. Image coding for machines: an end-to-end learned approach. In Proceedings of the IEEE International Conference on Acoustics, Speech and Signal Processing, 1590--1594

    2021

  16. [24]

    Li, B.; Ye, L.; Liang, J.; Wang, Y.; and Han, J. 2022. Region-of-interest and channel attention-based joint optimization of image compression and computer vision. Neurocomputing, 500: 13--25

    2022

  17. [25]

    Lin, T.-Y.; Maire, M.; Belongie, S.; Hays, J.; Perona, P.; Ramanan, D.; Doll \'a r, P.; and Zitnick, C. L. 2014. Microsoft coco: Common objects in context. In Proceedings of the European Conference on Computer Vision, 740--755

    2014

  18. [26]

    Liu, D.; Zhang, H.; and Xiong, Z. 2019. On the classification-distortion-perception tradeoff. In Proceedings of the International Conference on Neural Information Processing Systems, 1206--1215

    2019

  19. [27]

    Liu, J.; Sun, H.; and Katto, J. 2022. Improving multiple machine vision tasks in the compressed domain. In Proceedings of the IEEE International Conference on Pattern Recognition, 331--337

    2022

  20. [28]

    Liu, J.; Sun, H.; and Katto, J. 2023. Learned image compression with mixed transformer-cnn architectures. In Proceedings of the IEEE on Conference on Computer Vision and Pattern Recognition, 14388--14397

    2023

  21. [29]

    Liu, K.; Liu, D.; Li, L.; Yan, N.; and Li, H. 2021. Semantics-to-signal scalable image compression with learned revertible representations. International Journal of Computer Vision, 129(9): 2605--2621

    2021

  22. [30]

    Liu, L.; Hu, Z.; Chen, Z.; and Xu, D. 2023. Icmh-net: Neural image compression towards both machine vision and human vision. In Proceedings of the ACM International Conference on Multimedia, 8047--8056

    2023

  23. [31]

    Minnen, D.; Ball \'e , J.; and Toderici, G. D. 2018. Joint autoregressive and hierarchical priors for learned image compression. Advances in Neural Information Processing Systems, 31

    2018

  24. [32]

    Minnen, D.; and Singh, S. 2020. Channel-wise autoregressive entropy models for learned image compression. In Proceedings of the IEEE International Conference on Image Processing, 3339--3343

    2020

  25. [33]

    W.; Hallacy, C.; Ramesh, A.; Goh, G.; Agarwal, S.; Sastry, G.; Askell, A.; Mishkin, P.; Clark, J.; et al

    Radford, A.; Kim, J. W.; Hallacy, C.; Ramesh, A.; Goh, G.; Agarwal, S.; Sastry, G.; Askell, A.; Mishkin, P.; Clark, J.; et al. 2021. Learning transferable visual models from natural language supervision. In Proceedings of the International Conference on Machine Learning, 8748--8763

    2021

  26. [34]

    Redmon, J.; and Farhadi, A. 2018. Yolov3: An incremental improvement. arXiv preprint arXiv:1804.02767

    2018 arXiv

  27. [35]

    Shen, X.; Yin, K.; Wang, X.; He, Y.; Wang, S.; and Yang, W. 2024. Image Coding for Analytics via Adversarially Augmented Adaptation. In Proceedings of the IEEE International Conference on Acoustics, Speech and Signal Processing, 3605--3609

    2024

  28. [36]

    Shindo, T.; Yamada, K.; Watanabe, T.; and Watanabe, H. 2024. Image Coding for Machines with Edge Information Learning Using Segment Anything. arXiv preprint arXiv:2403.04173

    2024 arXiv

  29. [37]

    J.; Ohm, J.-R.; Han, W.-J.; and Wiegand, T

    Sullivan, G. J.; Ohm, J.-R.; Han, W.-J.; and Wiegand, T. 2012. Overview of the high efficiency video coding ( HEVC ) standard. IEEE Transactions on Circuits and Systems for Video technology, 22(12): 1649--1668

    2012

  30. [38]

    Torfason, R.; Mentzer, F.; Agustsson, E.; Tschannen, M.; Timofte, R.; and Van Gool, L. 2018. Towards image understanding from deep compression without decoding. arXiv preprint arXiv:1803.06131

    2018 arXiv

  31. [39]

    Wang, S.; Wang, S.; Yang, W.; Zhang, X.; Wang, S.; Ma, S.; and Gao, W. 2021 a . Towards analysis-friendly face representation with scalable feature and texture compression. IEEE Transactions on Multimedia, 24: 3169--3181

    2021

  32. [40]

    Wang, S.; Wang, Z.; Wang, S.; and Ye, Y. 2021 b . End-to-end compression towards machine vision: Network architecture design and optimization. IEEE Open Journal of Circuits and Systems, 2: 675--685

    2021

  33. [41]

    Wang, X.; Yu, Z.; De Mello, S.; Kautz, J.; Anandkumar, A.; Shen, C.; and Alvarez, J. M. 2022. Freesolo: Learning to segment objects without annotations. In Proceedings of the IEEE Conference on Computer Vision and Pattern Recognition, 14176--14186

    2022

  34. [42]

    Xie, Y.; Cheng, K.; and Chen, Q. 2021. Enhanced invertible encoding for learned image compression. In Proceedings of the ACM International Conference on Multimedia, 162--170

    2021

  35. [43]

    Yang, S.; Hu, Y.; Yang, W.; Duan, L.-Y.; and Liu, J. 2021. Towards coding for human and machine vision: Scalable face image coding. IEEE Transactions on Multimedia, 23: 2957--2971

    2021

  36. [44]

    Yang, W.; Huang, H.; Hu, Y.; Duan, L.; and Liu, J. 2024. Video Coding for Machines: Compact Visual Representation Compression for Intelligent Collaborative Analytics. IEEE Transactions on Pattern Analysis amp; Machine Intelligence, 46(07): 5174--5191

    2024

  37. [45]

    A.; Shechtman, E.; and Wang, O

    Zhang, R.; Isola, P.; Efros, A. A.; Shechtman, E.; and Wang, O. 2018. The unreasonable effectiveness of deep features as a perceptual metric. In Proceedings of the IEEE Conference on Computer Vision and Pattern Recognition, 586--595

    2018

  38. [46]

    H.; Zhou, L.; Dai, X.; Yuan, L.; Li, Y.; et al

    Zhong, Y.; Yang, J.; Zhang, P.; Li, C.; Codella, N.; Li, L. H.; Zhou, L.; Dai, X.; Yuan, L.; Li, Y.; et al. 2022. Regionclip: Region-based language-image pretraining. In Proceedings of the IEEE Conference on Computer Vision and Pattern Recognition, 16793--16803

    2022

  39. [47]

    , " * write output.state after.block = add.period write newline

    ENTRY address archivePrefix author booktitle chapter edition editor eid eprint howpublished institution isbn journal key month note number organization pages publisher school series title type volume year label extra.label sort.label short.list INTEGERS output.state before.all...

  40. [48]

    write newline

    " write newline "" before.all 'output.state := FUNCTION n.dashify 't := "" t empty not t #1 #1 substring "-" = t #1 #2 substring "--" = not "--" * t #2 global.max substring 't := t #1 #1 substring "-" = "-" * t #2 global.max substring 't := while if t #1 #1 substring * t #2 gl...

Pith tools

Reviewed August 10, 2026 · model on record in the stance chip above.