Pith. sign in

REVIEW 3 major objections 5 minor 1 cited by

Rethinking High-speed Image Reconstruction Framework with Spike Camera

T0 review · 3 major / 5 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 SpikeCLIP trains low-light spike-to-image reconstruction from CLIP text and unpaired high-quality images, and reports better reconstruction quality on U-CALTECH and U-CIFAR than prior methods.

desk verdict Novel CLIP-based training paradigm for spike reconstruction, undermined by an evaluation loop where NIQE selects the training targets and then scores the results. read the letter →

arxiv 2501.04477 v2 pith:HTIDWAPY submitted 2025-01-08 cs.CV

Kang Chen , Yajing Zheng , Tiejun Huang , Zhaofei Yu This is my paper · ORCID
classification cs.CV
keywords spikecameraimagereconstructionlow-lightimagingCLIPpromptlearningcontrastiveneuromorphicvisionno-referencequality
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

SpikeCLIP aims to show that a spike-to-image network can be trained for low-light scenes without any paired sharp ground truth, using only the class label of the scene and an unpaired collection of high-quality images. The paper argues that CLIP's shared text-image space turns these weak signals into a strong training signal, so a lightweight 0.186M-parameter network can reconstruct texture-rich, brightness-balanced images. This matters because previous supervised and self-supervised methods degrade on real low-light spikes, where synthetic training pairs do not match the noise and pseudo-labels lose quality. If the claim holds, class labels plus unpaired high-quality photos are sufficient supervision for practical spike-camera reconstruction under extreme lighting.

What carries the argument

The load-bearing mechanism is CLIP as a joint text-image embedding space, used as the only perceptual supervisor. Two learnable prompts, initialized in the style of CoOp, are optimized to represent the distributions of high-quality and low-quality reconstructions; the reconstruction network is then trained with a decoupled class loss and prompt loss in CLIP's shared embedding space. A supporting mechanism is the HQ-image pipeline, which merges outputs of several reconstruction algorithms (TFP, TFI, Spk2ImgNet, SSML, WGSE) and keeps the lowest-NIQE image per sample, while a TFI-based coarse reconstruction gives the lightweight network a stable initialization.

What would settle it

Take a held-out set of real low-light spike streams not used in training, reconstruct with SpikeCLIP and with WGSE or TFI, and run a forced-choice human perceptual study with equalized display. If human raters prefer SpikeCLIP no more often than chance, or if a perceptual metric that was not used to build the HQ dataset (for example, a trained opinion-aware image quality model) shows no advantage, the claimed perceptual improvement would not hold.

Watch

Extended reading notes

Core claim

On its own terms, the paper's central claim is that SpikeCLIP surpasses previous spike-to-image reconstruction methods on the real-world low-light datasets U-CALTECH and U-CIFAR, and that it does so by replacing ground-truth supervision with two CLIP-based losses. The method has three stages: a coarse stage trains a lightweight reconstruction network to output the model-based TFI estimate; a prompt-learning stage builds a high-quality image dataset from synthetic and real reconstructions and learns CLIP prompts that separate high-quality from low-quality images; and a fine stage optimizes the network with a class loss (InfoNCE against the class label's text feature) plus a prompt loss that pulls reconstructed images toward the high-quality prompt. The paper reports NIQE reductions of 37.9% on U-CALTECH and 43.1% on U-CIFAR over the best compared method, along with lower BRISQUE and PIQE scores, a 64% classification accuracy on U-CALTECH, and a 0.186M-parameter network with 0.540 ms latency.

Load-bearing premise

The evaluation assumes that NIQE and the other no-reference metrics measure true reconstruction quality, even though the training's high-quality dataset was built by selecting the lowest-NIQE image, so the reported gains may partly reflect optimizing that same statistic.

Editorial extensions

If this is right

  • If the claim holds, low-light spike-to-image reconstruction no longer requires paired sharp ground truth; class labels plus unpaired high-quality images suffice, removing the synthetic-to-real domain gap as a training obstacle.
  • A 0.186M-parameter network with 0.540 ms latency is enough to beat much larger supervised models, suggesting CLIP supervision can make low-power spike cameras practical for edge deployment.
  • Because outputs are aligned with CLIP's text-image features, reconstructed images can be directly classified or matched to text without a separate recognition head.
  • On U-CALTECH and U-CIFAR, the method reports the best NIQE, BRISQUE, and PIQE among the compared methods, with the reported margins shown in the results table.

Reading between the lines

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

  • Editorial inference: because the supervision is purely semantic, the same recipe should transfer to other degradation types such as motion blur, haze, or extreme noise wherever a CLIP-recognizable label exists; a direct test would be applying SpikeCLIP to normal-light or deblurring spike datasets.
  • Editorial inference: the HQ dataset is built by minimizing NIQE, and the headline metrics are also NIQE-family, so a cleaner test of perceptual quality would use a held-out human rating study or a metric not involved in dataset construction.
  • Editorial inference: the reported classification accuracy suggests the reconstruction network is learning semantic features from spikes even before image quality is perfect, hinting that joint reconstruction-recognition systems could be trained end-to-end with the same losses, a direction the paper does not develop.
Share X Bluesky LinkedIn Reddit HN

Signed reviews

No signed human review yet.

Editorial analysis

A structured set of objections, weighed in public.

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

Referee Report

3 major / 5 minor

Summary. The paper proposes SpikeCLIP, a three-stage spike-to-image reconstruction framework for real-world low-light spike streams. Stage 1 trains a lightweight reconstruction network to reproduce TFI results; Stage 2 constructs a high-quality (HQ) image set from synthetic and real datasets, selecting candidate reconstructions with the lowest NIQE score (Eq. 6), and learns prompt embeddings that separate HQ from low-quality images in CLIP space; Stage 3 fine-tunes the network with a class-label InfoNCE loss (Eq. 10) and a prompt loss (Eq. 9). Experiments on U-CALTECH and U-CIFAR report large improvements over prior methods in NIQE, BRISQUE, and PIQE, higher CLIP classification accuracy, and very low parameter count, FLOPs, and latency. The central claim is that this CLIP-based supervision, using class labels and unpaired HQ images, suffices to train a lightweight network that recovers texture-rich and brightness-balanced images under low light without paired ground truth.

Significance. If the quantitative claims hold, the framework is a meaningful contribution: it removes the requirement of paired spike-sharp supervision in real-world low-light spike reconstruction, exploits CLIP's text-image alignment as a supervisory signal, and demonstrates that a lightweight network can suffice when the supervision is semantic. The paper also releases code, which supports reproducibility. However, the evaluation protocol is too weak to support the central claim as it stands: the training-target selection and the headline evaluation metric are not independent, and the method uses class labels that the baselines do not. The core idea is promising, but the evidence for superior reconstruction quality requires independent validation.

major comments (3)
  1. [Prompt Learning, Eq. (6) and Experimental Results, Table 1] The evaluation of reconstruction quality is circular with respect to the HQ dataset construction. In Eq. (6), the HQ images used as positive training targets are selected by minimizing NIQE among candidate reconstructions on real-world datasets, and the main results in Table 1 are reported as NIQE improvements. Because the prompt-loss and fine-tuning stages align the LRN output with embeddings of these NIQE-selected images, a large part of the reported NIQE gain is expected by construction rather than being evidence of genuine perceptual improvement. The claim that SpikeCLIP 'significantly enhances texture details and luminance balance' is therefore not established by Table 1. Please provide evaluation with a metric not involved in dataset construction, such as a human perceptual study, or report scores on an HQ set selected without NIQE.
  2. [Fine Reconstruction, Eq. (10), and Table 1] The comparison with prior methods is unequal because SpikeCLIP is trained with ground-truth class labels for the reconstructed scene, whereas TFP, TFI, TFSTP, SSML, SpikeFormer, RSIR, Spk2ImgNet, and WGSE do not use label information. The gains in Table 1 may reflect the additional semantic supervision rather than a fundamentally better spike-to-image mapping. In particular, the classification accuracy reported in Table 2 measures alignment with the text labels, not image reconstruction quality. Please compare against baselines that also receive the same class-label information, or explicitly justify why the comparison remains fair for the reconstruction-quality claim.
  3. [Experimental Results, Table 1 and Fig. 5] All quantitative quality evidence rests on no-reference metrics NIQE, BRISQUE, and PIQE, which are known to correlate imperfectly with perceptual quality and can be optimized artificially. There is no real-world paired ground truth, and Fig. 5 shows only selected qualitative examples. Please report variability across the 1,000 test samples, include statistical significance tests and error bars, and provide a more complete qualitative panel rather than a single selected frame per scene. This would substantially increase confidence that the improvements are perceptual and not metric-specific.
minor comments (5)
  1. [Fig. 5 caption] The caption contains a typo: 'U-CALTEHC' should be 'U-CALTECH'.
  2. [Eq. (9)] The prompt loss in Eq. (9) is the negative softmax probability without a logarithm; please clarify whether this is intentional or whether a cross-entropy formulation was intended, since the loss magnitude and gradient scale differ.
  3. [Table 3] The columns 'Params' and 'Flops' in Table 3 appear to omit units; they are presumably millions and giga-FLOPs, respectively, and should be labeled consistently with Table 1.
  4. [Eq. (6)] The text says the image with the 'best non-reference metric NIQE score' is selected, but because lower NIQE is better, 'best' should be 'lowest' for clarity.
  5. [Table 1] The text says bold and underline distinguish best and second-best results, but Table 1 shows only bold entries and no underlined entries; please align the formatting with the stated policy.

Circularity Check

1 steps flagged · score 6.0 of 10

Reported NIQE gains are partly circular: HQ training targets are selected by minimizing NIQE and the same NIQE metric is used as the headline evaluation.

  1. fitted input called prediction [Section 'Prompt Learning / HQ Images Generation Pipeline', Eq. (6); Section 'Experiments / Training Details', evaluation description; Table 1]
    "Among them, image Ihq with the best non-reference metric NIQE (Mittal, Soundararajan, and Bovik 2012) score is added to the HQ-Images dataset, i.e.: Ihq = arg min_{m in M} NIQE(Im) (6) ... Since paired spike-sharp datasets are not available in real-world scenarios, we use non-reference metrics, NIQE ..., BRISQUE ..., and PIQE ..., to evaluate the image reconstruction quality."

    The supervision dataset is constructed by selecting, per scene, the reconstruction with the lowest NIQE among TFP, TFI, Spk2ImgNet, SSML, and WGSE (Eq. 6). The fine-reconstruction prompt loss then trains the lightweight network to align its outputs with the CLIP embedding of these NIQE-selected HQ images. The paper's central quantitative claim of superiority is then reported in Table 1 using the same NIQE statistic. Thus part of the reported NIQE improvement is a selection effect: the model was guided toward images that were explicitly chosen to minimize NIQE, so NIQE is not an independent measure of the claimed perceptual gains.

full rationale

The derivation chain is not fully self-contained: the CLIP prompt supervision is anchored to a high-quality dataset whose construction criterion (minimum NIQE, Eq. 6) is identical to the headline evaluation metric in Table 1. This makes the NIQE column of Table 1 a partly fitted target rather than a fully independent prediction. However, the method also uses class labels and CLIP text features as external supervision, and it improves on BRISQUE and PIQE, which were not used in the HQ selection. The LRN architecture is borrowed from prior work and the framework is tested against external baselines on U-CALTECH/U-CIFAR, so the central claim has independent content beyond the NIQE-selection bias. No self-citation chain is load-bearing here, and no uniqueness theorem is imported. The main circularity risk is confined to the NIQE evaluation channel; it does not make the entire reconstruction result an equivalence by definition.

Assumptions & free parameters 3 free parameters · 5 assumptions · 0 invented entities

The method adds no new physical entities. It rests on domain assumptions about the spike camera model, TFI as a coarse target, CLIP as a quality-aware supervisor, and NIQE as a trustworthy metric. The main free hyperparameters are the prompt loss weight, the InfoNCE temperature, and the voxelization length; the first is set to 100 without sensitivity analysis.

free parameters (3)
  • prompt loss weight lambda = 100
    Set by hand as the weighting of Lprompt in the total loss (Eq. 11). The paper does not report a sensitivity analysis, and this value affects the balance between semantic and prompt alignment.
  • InfoNCE temperature tau = not specified
    Appears in Eq. 10; its value is never stated, so the contrastive loss scale is an underspecified free choice.
  • voxelized spike length = 50 bins
    The spike stream length is reduced from 200 to 50 time bins before reconstruction. This is a hand-chosen preprocessing parameter that affects temporal information available to the network.
assumptions (5)
  • domain assumption The integrate-and-fire spike camera model in Eq. 1 describes the real capture process.
    The entire framing assumes spikes are emitted when accumulated intensity crosses a threshold; real devices may have noise and calibration mismatch.
  • domain assumption TFI provides a valid coarse reconstruction target for low-light spike streams.
    Coarse reconstruction is trained to regress TFI output, so if TFI is poor under low light, the initialization is poor.
  • domain assumption CLIP embeddings are sensitive enough to image quality to distinguish high-quality from low-quality reconstructions and to guide low-level restoration.
    The prompt loss and class loss both rely on CLIP text-image alignment transferring to low-level reconstruction quality, which is not guaranteed.
  • domain assumption NIQE is a reliable perceptual quality measure for spike-camera low-light images.
    NIQE is used both to select high-quality training images and to evaluate the final results, so the paper depends on NIQE tracking true perceptual quality.
  • domain assumption Class labels for captured scenes are available during training and match the CLIP vocabulary.
    U-CALTECH and U-CIFAR provide labels, but general deployment would require labels or text descriptions for arbitrary scenes.

how reviews work

0 comments
Cite this review

Pith. "Pith review of Rethinking High-speed Image Reconstruction Framework with Spike Camera." pith.science (2026). https://pith.science/paper/HTIDWAPY

@misc{pith2026250104477,
  author       = {Pith},
  title        = {Pith review of: Rethinking High-speed Image Reconstruction Framework with Spike Camera},
  year         = {2026},
  howpublished = {\url{https://pith.science/paper/HTIDWAPY}},
  note         = {Machine review of arXiv:2501.04477}
}
read the original abstract

Spike cameras, as innovative neuromorphic devices, generate continuous spike streams to capture high-speed scenes with lower bandwidth and higher dynamic range than traditional RGB cameras. However, reconstructing high-quality images from the spike input under low-light conditions remains challenging. Conventional learning-based methods often rely on the synthetic dataset as the supervision for training. Still, these approaches falter when dealing with noisy spikes fired under the low-light environment, leading to further performance degradation in the real-world dataset. This phenomenon is primarily due to inadequate noise modelling and the domain gap between synthetic and real datasets, resulting in recovered images with unclear textures, excessive noise, and diminished brightness. To address these challenges, we introduce a novel spike-to-image reconstruction framework SpikeCLIP that goes beyond traditional training paradigms. Leveraging the CLIP model's powerful capability to align text and images, we incorporate the textual description of the captured scene and unpaired high-quality datasets as the supervision. Our experiments on real-world low-light datasets U-CALTECH and U-CIFAR demonstrate that SpikeCLIP significantly enhances texture details and the luminance balance of recovered images. Furthermore, the reconstructed images are well-aligned with the broader visual features needed for downstream tasks, ensuring more robust and versatile performance in challenging environments.

Figures

Figures reproduced from arXiv: 2501.04477 by the authors.

Figure 1
Figure 1. Illustration of the advantages of our method. While previous learning-based approaches struggle with real-world data [PITH_FULL_IMAGE:figures/full_fig_p002_1.png] view at source ↗
Figure 2
Figure 2. The overall framework of our three-stage spike-based image reconstruction method. [PITH_FULL_IMAGE:figures/full_fig_p003_2.png] view at source ↗
Figure 3
Figure 3. The framework of our designed HQ images generation pipeline. [PITH_FULL_IMAGE:figures/full_fig_p004_3.png] view at source ↗
Figures from the paper (2 more)
Figure 4
Figure 4. Figure 4: Prompt and class loss illustration. Prompt Loss. With the learnable HQ and LQ prompts learned from the initialization stage, we can further optimize the LRN network like the previous study (Liang et al. 2023). Specifically, we design the prompt loss to measure the alig…
Figure 5
Figure 5. Figure 5: Visual comparison of our method with previous methods on the U-CALTEHC dataset. [PITH_FULL_IMAGE:figures/full_fig_p006_5.png]

Discussion (0). Continue with ORCID to comment.

Forward citations

Cited by 1 Pith paper

Reviewed papers in the Pith corpus that reference this work. Sorted by Pith novelty score. Full citation record

  1. USP-Gaussian: Unifying Spike-based Image Reconstruction, Pose Correction and Gaussian Splatting

    cs.CV 2024-11 conditional novelty 5.0 of 10

    USP-Gaussian jointly optimizes spike-to-image reconstruction, camera poses, and 3D Gaussian Splatting, reducing cascaded errors and improving 3D reconstruction quality on synthetic and real-world spike-camera data.

Reference graph

Works this paper leans on

33 extracted references · 27 canonical work pages · cited by 1 Pith paper

  1. [1]

    , " * 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 mid.sentence after.sentence after.block FUNCTION init.state.consts #0 'before.a...

  2. [2]

    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 global.max substring 't := if while FUNCTION word.in bbl.in capitalize " " * FUNCT...

  3. [3]

    Chen, K.; Chen, S.; Zhang, J.; Zhang, B.; Zheng, Y.; Huang, T.; and Yu, Z. 2024. SpikeReveal: Unlocking Temporal Sequences from Real Blurry Inputs with Spike Streams. arXiv preprint arXiv:2403.09486

    work page Pith review arXiv 2024

  4. [4]

    Chen, K.; Ye, Z.; Huang, T.; and Yu, Z. 2025. Spike-Zoo : A Toolbox for Spike-to-Image Reconstruction. https://github.com/chenkang455/Spike-Zoo

    work page 2025

  5. [5]

    Chen, K.; and Yu, L. 2024. Motion Deblur by Learning Residual from Events. IEEE TMM, 26: 6632--6647

    work page 2024

  6. [6]

    Chen, S.; Duan, C.; Yu, Z.; Xiong, R.; and Huang, T. 2022. Self-supervised mutual learning for dynamic scene reconstruction of spiking camera. IJCAI

    work page 2022

  7. [7]

    Chen, S.; Yu, Z.; and Huang, T. 2023. Self-supervised joint dynamic scene reconstruction and optical flow estimation for spiking camera. In AAAI, volume 37, 350--358

    work page 2023

  8. [8]

    Cheng, J.; Liang, D.; and Tan, S. 2024. Transfer CLIP for Generalizable Image Denoising. In CVPR, 25974--25984

    work page 2024

Show all 33 references
  1. [9]

    Cho, H.; Kim, H.; Chae, Y.; and Yoon, K.-J. 2023. Label-free event-based object recognition via joint learning with image reconstruction from events. In ICCV, 19866--19877

    2023

  2. [10]

    Fei-Fei, L.; Fergus, R.; and Perona, P. 2004. Learning generative visual models from few training examples: An incremental bayesian approach tested on 101 object categories. In CVPR, 178--178. IEEE

    2004

  3. [11]

    Huang, T.; Zheng, Y.; Yu, Z.; Chen, R.; Li, Y.; Xiong, R.; Ma, L.; Zhao, J.; Dong, S.; Zhu, L.; et al. 2023. 1000 faster camera and machine vision with ordinary devices. Engineering, 25: 110--119

    2023

  4. [12]

    Krizhevsky, A.; Hinton, G.; et al. 2009. Learning multiple layers of features from tiny images

    2009

  5. [13]

    Liang, Z.; Li, C.; Zhou, S.; Feng, R.; and Loy, C. C. 2023. Iterative prompt learning for unsupervised backlit image enhancement. In ICCV, 8094--8103

    2023

  6. [14]

    K.; and Bovik, A

    Mittal, A.; Moorthy, A. K.; and Bovik, A. C. 2012. No-reference image quality assessment in the spatial domain. TIP, 21(12): 4695--4708

    2012

  7. [15]

    completely blind

    Mittal, A.; Soundararajan, R.; and Bovik, A. C. 2012. Making a “completely blind” image quality analyzer. SPL, 20(3): 209--212

    2012

  8. [16]

    Oord, A. v. d.; Li, Y.; and Vinyals, O. 2018. Representation learning with contrastive predictive coding. arXiv preprint arXiv:1807.03748

    2018 arXiv

  9. [17]

    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 ICML, 8748--8763. PMLR

    2021

  10. [18]

    Ronneberger, O.; Fischer, P.; and Brox, T. 2015. U-net: Convolutional networks for biomedical image segmentation. In Medical Image Computing and Computer-Assisted Intervention--MICCAI 2015: 18th International Conference, Munich, Germany, October 5-9, 2015, Proceedings, Part II...

    2015

  11. [19]

    She, C.; and Qing, L. 2022. SpikeFormer: Image Reconstruction from the Sequence of Spike Camera Based on Transformer. In Proceedings of the 2022 5th International Conference on Image and Graphics Processing, 72--78

    2022

  12. [20]

    Song, C.; Huang, Q.; and Bajaj, C. 2022. E-cir: Event-enhanced continuous intensity recovery. In CVPR, 7803--7812

    2022

  13. [21]

    C.; Channappayya, S

    Venkatanath, N.; Praneeth, D.; Bh, M. C.; Channappayya, S. S.; and Medasani, S. S. 2015. Blind image quality evaluation using perception based features. In 2015 twenty first national conference on communications (NCC), 1--6. IEEE

    2015

  14. [22]

    Yang, H.; Pan, L.; Yang, Y.; Hartley, R.; and Liu, M. 2024. LDP: Language-driven Dual-Pixel Image Defocus Deblurring Network. In CVPR, 24078--24087

    2024

  15. [23]

    Zhang, J.; Jia, S.; Yu, Z.; and Huang, T. 2023. Learning temporal-ordered representation for spike streams based on discrete wavelet transforms. In AAAI, volume 37, 137--147

    2023

  16. [24]

    Zhao, J.; Xiong, R.; Liu, H.; Zhang, J.; and Huang, T. 2021. Spk2imgnet: Learning to reconstruct dynamic scene from continuous spike stream. In CVPR, 11996--12005

    2021

  17. [25]

    Zhao, J.; Zhang, S.; Yu, Z.; and Huang, T. 2024 a . Recognizing Ultra-High-Speed Moving Objects with Bio-Inspired Spike Camera. In AAAI, volume 38, 7478--7486

    2024

  18. [26]

    Zhao, R.; Xiong, R.; Zhao, J.; Zhang, J.; Fan, X.; Yu, Z.; and Huang, T. 2024 b . Boosting Spike Camera Image Reconstruction from a Perspective of Dealing with Spike Fluctuations. In CVPR, 24955--24965

    2024

  19. [27]

    Zheng, Y.; Yu, Z.; Wang, S.; and Huang, T. 2022. Spike-based motion estimation for object tracking through bio-inspired unsupervised learning. TIP, 32: 335--349

    2022

  20. [28]

    Zheng, Y.; Zhang, J.; Zhao, R.; Ding, J.; Chen, S.; Xiong, R.; Yu, Z.; and Huang, T. 2023. SpikeCV: Open a Continuous Computer Vision Era. arXiv preprint arXiv:2303.11684

    2023 arXiv

  21. [29]

    Zheng, Y.; Zheng, L.; Yu, Z.; Shi, B.; Tian, Y.; and Huang, T. 2021. High-speed image reconstruction through short-term plasticity for spiking cameras. In CVPR, 6358--6367

    2021

  22. [30]

    C.; and Liu, Z

    Zhou, K.; Yang, J.; Loy, C. C.; and Liu, Z. 2022. Learning to prompt for vision-language models. IJCV, 130(9): 2337--2348

    2022

  23. [31]

    Zhu, L.; Dong, S.; Huang, T.; and Tian, Y. 2019. A retina-inspired sampling method for visual texture reconstruction. In ICME, 1432--1437. IEEE

    2019

  24. [32]

    Zhu, L.; Dong, S.; Li, J.; Huang, T.; and Tian, Y. 2020. Retina-like visual image reconstruction via spiking neural model. In CVPR, 1438--1446

    2020

  25. [33]

    Zhu, L.; Zheng, Y.; Geng, M.; Wang, L.; and Huang, H. 2023. Recurrent spike-based image restoration under general illumination. In ACMMM, 8251--8260

    2023

Pith tools

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