pith. sign in

arxiv: 2511.17183 · v2 · submitted 2025-11-21 · 💻 cs.CV · cs.CY

Learning Under Low Illumination: A Dataset and Algorithm for Traffic Sign Recognition

Aditya Mishra , Akshay Agarwal , Haroon Lone This is my paper

Pith reviewed 2026-05-17 20:37 UTC · model grok-4.3

classification 💻 cs.CV cs.CY
keywords nighttime traffic sign recognitionlow-light datasetINTSDLENS-Netobject detectionautonomous drivingillumination-aware models
0
0 comments X

The pith

A new Indian nighttime traffic sign dataset and LENS-Net model tackle recognition under low illumination.

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

The paper seeks to close the gap in traffic sign recognition for nighttime driving by releasing a large realistic dataset and a dedicated baseline algorithm. Daytime datasets miss real degradations such as headlight glare, motion blur, sensor noise, and ambiguous or vandalized signs. INTSD supplies thousands of street-level images spanning 41 sign classes plus distractor categories collected across varied Indian regions, lighting, and weather. LENS-Net combines adaptive illumination-aware detection with multimodal semantic reasoning to produce more reliable classifications than standard models on these images. The result supplies standardized benchmarks and a public starting point for algorithms that must work when visibility drops.

Core claim

We introduce INTSD, a large-scale nighttime traffic sign dataset collected across diverse regions of India containing street-level images spanning 41 traffic signboard classes, multiple distractor categories, and varied lighting and weather conditions. We present LENS-Net, which integrates adaptive illumination-aware detection with multimodal semantic reasoning for robust nighttime sign classification, and show through evaluations that current models face clear challenges on this data.

What carries the argument

LENS-Net, which integrates adaptive illumination-aware detection with multimodal semantic reasoning to classify signs despite low-light degradations.

If this is right

  • Standard daytime detectors and classifiers will underperform on INTSD, confirming the need for illumination-specific methods.
  • LENS-Net provides a concrete baseline that future nighttime recognition work can compare against.
  • The dataset enables joint training for detection and fine-grained classification under realistic low-light conditions.
  • Autonomous driving and intelligent transportation systems can now be tested against documented nighttime sign challenges.

Where Pith is reading between the lines

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

  • The same illumination-aware and multimodal techniques might transfer to other low-light vision tasks such as pedestrian or vehicle detection.
  • Extending the collection protocol to additional countries or sensor types would test whether the observed performance gaps are universal.
  • Integration of the dataset with existing daytime benchmarks could produce training regimes that maintain accuracy across day-to-night transitions.

Load-bearing premise

The collected images and labeling process accurately capture the full range of real-world low-light degradations and distractors that future systems will encounter outside the Indian collection sites.

What would settle it

A detection or classification model that reaches high accuracy on INTSD but shows large drops in performance on nighttime traffic signs collected from a different country or under lighting conditions absent from the dataset would show the data and model do not fully generalize.

Figures

Figures reproduced from arXiv: 2511.17183 by Aditya Mishra, Akshay Agarwal, Haroon Lone.

Figure 1
Figure 1. Figure 1: Geographical coverage and representative nighttime traf [PITH_FULL_IMAGE:figures/full_fig_p003_1.png] view at source ↗
Figure 2
Figure 2. Figure 2: The treemap (on left) illustrates the class distribution. The bar graphs (on right) gives distribution of image size (golden) and [PITH_FULL_IMAGE:figures/full_fig_p004_2.png] view at source ↗
Figure 3
Figure 3. Figure 3: Overview of the proposed LENS-Net architecture. The LENS-Net detector (top) integrates adaptive image enhancement for joint [PITH_FULL_IMAGE:figures/full_fig_p005_3.png] view at source ↗
Figure 4
Figure 4. Figure 4: Fold-wise performance of LENS-Net for detection (left) and classification (right) on INTSD. A.P., A.R. and Acc. denotes macro [PITH_FULL_IMAGE:figures/full_fig_p007_4.png] view at source ↗
Figure 5
Figure 5. Figure 5: Qualitative results of LENS-Net on INTSD. The green, red, and blue boxes indicate true positives, false negatives, and false [PITH_FULL_IMAGE:figures/full_fig_p008_5.png] view at source ↗
Figure 6
Figure 6. Figure 6: Ablation study on the impact of major components in [PITH_FULL_IMAGE:figures/full_fig_p008_6.png] view at source ↗
Figure 7
Figure 7. Figure 7: Examples of dataset images from smartphone (left) [PITH_FULL_IMAGE:figures/full_fig_p012_7.png] view at source ↗
Figure 8
Figure 8. Figure 8: Illustration of selected challenges within the INTSD, [PITH_FULL_IMAGE:figures/full_fig_p012_8.png] view at source ↗
Figure 12
Figure 12. Figure 12: Images with multiple signboards in one frame, creating [PITH_FULL_IMAGE:figures/full_fig_p013_12.png] view at source ↗
Figure 13
Figure 13. Figure 13: Four distinct signboard designs, all conveying mapping [PITH_FULL_IMAGE:figures/full_fig_p013_13.png] view at source ↗
Figure 11
Figure 11. Figure 11: Images captured under low-light conditions, illustrating [PITH_FULL_IMAGE:figures/full_fig_p013_11.png] view at source ↗
Figure 15
Figure 15. Figure 15: Illustrative samples of images labeled as [PITH_FULL_IMAGE:figures/full_fig_p014_15.png] view at source ↗
Figure 18
Figure 18. Figure 18: Sample instances of the day images from INTSD. The [PITH_FULL_IMAGE:figures/full_fig_p014_18.png] view at source ↗
read the original abstract

Traffic signboards are vital for road safety and intelligent transportation systems, enabling navigation and autonomous driving. Yet, recognizing traffic signs at night remains underexplored due to the scarcity of realistic public datasets capturing low-light degradations and distractor classes. Existing benchmarks are predominantly daytime and do not reflect challenges such as headlight glare, motion blur, sensor noise, and vandalized or ambiguous signage. To address these gaps, we introduce INTSD, a large-scale nighttime traffic sign dataset collected across diverse regions of India. INTSD contains street-level images spanning 41 traffic signboard classes, multiple distractor categories, and varied lighting and weather conditions. The dataset is designed to support both detection and fine-grained classification under realistic nighttime scenarios. To benchmark INTSD for nighttime sign recognition, we conduct extensive evaluations using state-of-the-art detection and classification models under standardized protocols. Additionally, we present LENS-Net, a strong baseline that integrates adaptive illumination-aware detection with multimodal semantic reasoning for robust nighttime sign classification. Experiments and ablations demonstrate the challenges posed by INTSD and establish competitive baselines for future research. The dataset and code for LENS-Net is publicly available for research.

Editorial analysis

A structured set of objections, weighed in public.

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

Referee Report

2 major / 1 minor

Summary. The manuscript introduces INTSD, a large-scale nighttime traffic sign dataset collected across diverse regions of India containing street-level images spanning 41 traffic sign classes, multiple distractor categories, and varied lighting/weather conditions. It also presents LENS-Net as a baseline integrating adaptive illumination-aware detection with multimodal semantic reasoning, reports extensive evaluations of SOTA detection and classification models on the dataset under standardized protocols, and releases the dataset and code publicly.

Significance. If the dataset collection protocol and labeling accurately capture realistic low-light degradations and if LENS-Net evaluations are robust, the work would provide a valuable new benchmark and baseline for nighttime traffic sign recognition, addressing a documented scarcity in public datasets for this safety-critical application in intelligent transportation systems.

major comments (2)
  1. [§3 (Dataset Collection)] §3 (Dataset Collection): The abstract positions INTSD as addressing the global scarcity of realistic nighttime benchmarks capturing headlight glare, motion blur, sensor noise, and ambiguous signage. However, the collection is restricted to diverse regions of India; no discussion or validation addresses potential systematic differences in traffic sign standards, road infrastructure, headlight spectra, or extreme weather outside India. This directly affects whether INTSD can serve as a generalizable benchmark supporting the central claims.
  2. [§5 (Experiments and Ablations)] §5 (Experiments and Ablations): The abstract reports extensive evaluations and ablations demonstrating challenges and establishing competitive baselines for LENS-Net, yet no error bars, statistical significance tests, or details on post-hoc model selection choices are referenced. This makes it difficult to assess the strength of claims that LENS-Net provides a strong baseline relative to SOTA models.
minor comments (1)
  1. [Abstract] The abstract and introduction could include the total number of images and annotation statistics for INTSD to better convey dataset scale upfront.

Simulated Author's Rebuttal

2 responses · 0 unresolved

We thank the referee for their thoughtful review and constructive comments on our manuscript. We address each of the major comments point by point below, indicating the revisions we intend to make to improve the paper.

read point-by-point responses

  1. Referee: [§3 (Dataset Collection)] The abstract positions INTSD as addressing the global scarcity of realistic nighttime benchmarks capturing headlight glare, motion blur, sensor noise, and ambiguous signage. However, the collection is restricted to diverse regions of India; no discussion or validation addresses potential systematic differences in traffic sign standards, road infrastructure, headlight spectra, or extreme weather outside India. This directly affects whether INTSD can serve as a generalizable benchmark supporting the central claims.

    Authors: We appreciate the referee pointing out the need for clearer discussion on the geographic scope of INTSD. The dataset was specifically collected in India to capture realistic nighttime conditions in a region with diverse road infrastructure and sign variations that are underrepresented in existing benchmarks. However, we agree that without explicit discussion, it may be unclear how well the results generalize. In the revised manuscript, we will add a dedicated paragraph in Section 3 discussing potential systematic differences, such as variations in traffic sign designs across countries, differences in headlight technologies, and weather extremes not present in our collection sites. We will also note that while INTSD provides a valuable benchmark for low-illumination traffic sign recognition, cross-dataset validation with other regions would be beneficial for broader claims. This addition will help readers understand the intended scope and limitations. revision: yes

  2. Referee: [§5 (Experiments and Ablations)] The abstract reports extensive evaluations and ablations demonstrating challenges and establishing competitive baselines for LENS-Net, yet no error bars, statistical significance tests, or details on post-hoc model selection choices are referenced. This makes it difficult to assess the strength of claims that LENS-Net provides a strong baseline relative to SOTA models.

    Authors: We thank the referee for this observation on the experimental reporting. Our evaluations were conducted over multiple random seeds to ensure reliability, but we acknowledge that the manuscript does not explicitly include error bars or statistical tests in the presented results. To address this, we will revise Section 5 to include standard deviation error bars in all performance tables and figures. Furthermore, we will add statistical significance testing (such as McNemar's test for classification or paired t-tests for detection metrics) between LENS-Net and the compared SOTA models. We will also elaborate on the hyperparameter tuning and model selection procedure to provide full transparency. These revisions will strengthen the evidence for LENS-Net as a competitive baseline. revision: yes

Circularity Check

0 steps flagged

No circularity: contributions are new data collection and architecture proposal

full rationale

The paper introduces a new nighttime traffic sign dataset (INTSD) collected in India and proposes LENS-Net as a baseline model integrating illumination-aware detection and multimodal reasoning. No load-bearing steps reduce by construction to fitted parameters, self-citations, or prior ansatzes. Evaluations use standard SOTA models on the new data under explicit protocols; the dataset itself is the primary input rather than a derived output. The central claims rest on empirical collection and architectural design, which are self-contained against external benchmarks and do not invoke uniqueness theorems or self-referential definitions.

Axiom & Free-Parameter Ledger

0 free parameters · 1 axioms · 1 invented entities

The paper rests on standard computer-vision assumptions about labeled image data and the ability of neural networks to learn illumination-invariant features; no major free parameters or invented physical entities are introduced beyond the model name itself.

axioms (1)
  • domain assumption Deep neural networks trained on labeled images can learn features that generalize to unseen low-light conditions when sufficient diverse data is provided.
    Implicit in the benchmarking and baseline design described in the abstract.
invented entities (1)
  • LENS-Net no independent evidence
    purpose: Baseline model combining adaptive illumination-aware detection with multimodal semantic reasoning.
    Newly proposed architecture whose performance is demonstrated only on the paper's own experiments.

pith-pipeline@v0.9.0 · 5507 in / 1337 out tokens · 33109 ms · 2026-05-17T20:37:31.954360+00:00 · methodology

discussion (0)

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

Lean theorems connected to this paper

Citations machine-checked in the Pith Canon. Every link opens the source theorem in the public Lean library.

What do these tags mean?
matches
The paper's claim is directly supported by a theorem in the formal canon.
supports
The theorem supports part of the paper's argument, but the paper may add assumptions or extra steps.
extends
The paper goes beyond the formal theorem; the theorem is a base layer rather than the whole result.
uses
The paper appears to rely on the theorem as machinery.
contradicts
The paper's claim conflicts with a theorem or certificate in the canon.
unclear
Pith found a possible connection, but the passage is too broad, indirect, or ambiguous to say the theorem truly supports the claim.

Reference graph

Works this paper leans on

62 extracted references · 62 canonical work pages · 1 internal anchor

  1. [1]

    An erudite fine-grained visual classification model

    Dongliang Chang, Yujun Tong, Ruoyi Du, Timothy Hospedales, Yi-Zhe Song, and Zhanyu Ma. An erudite fine-grained visual classification model. InProceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition, pages 7268–7277, 2023. 2

    work page 2023

  2. [2]

    Learning to see in the dark

    Chen Chen, Qifeng Chen, Jia Xu, and Vladlen Koltun. Learning to see in the dark. InProceedings of the IEEE con- ference on computer vision and pattern recognition, pages 3291–3300, 2018. 1

    work page 2018

  3. [3]

    Junzhou Chen, Heqiang Huang, Ronghui Zhang, Nengchao Lyu, Yanyong Guo, Hong-Ning Dai, and Hong Yan. Yolo-ts: Real-time traffic sign detection with enhanced accuracy us- 8 ing optimized receptive fields and anchor-free fusion.IEEE Transactions on Intelligent Transportation Systems, 2025. 7

    work page 2025

  4. [4]

    You only need 90k parameters to adapt light: a light weight trans- former for image enhancement and exposure correction

    Ziteng Cui, Kunchang Li, Lin Gu, Shenghan Su, Peng Gao, Zhengkai Jiang, Yu Qiao, and Tatsuya Harada. You only need 90k parameters to adapt light: a light weight trans- former for image enhancement and exposure correction. arXiv preprint arXiv:2205.14871, 2022. 2

    work page arXiv 2022

  5. [5]

    Superpixels and graph con- volutional neural networks for efficient detection of nutri- ent deficiency stress from aerial imagery

    Saba Dadsetan, David Pichler, David Wilson, Naira Hov- akimyan, and Jennifer Hobbs. Superpixels and graph con- volutional neural networks for efficient detection of nutri- ent deficiency stress from aerial imagery. InProceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition, pages 2950–2959, 2021. 2

    work page 2021

  6. [6]

    R-fcn: Object detection via region-based fully convolutional networks.Ad- vances in neural information processing systems, 29, 2016

    Jifeng Dai, Yi Li, Kaiming He, and Jian Sun. R-fcn: Object detection via region-based fully convolutional networks.Ad- vances in neural information processing systems, 29, 2016. 2

    work page 2016

  7. [7]

    Boosting object detection with zero-shot day-night domain adaptation

    Zhipeng Du, Miaojing Shi, and Jiankang Deng. Boosting object detection with zero-shot day-night domain adaptation. InProceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition, pages 12666–12676, 2024. 7

    work page 2024

  8. [8]

    The via annota- tion software for images, audio and video

    Abhishek Dutta and Andrew Zisserman. The via annota- tion software for images, audio and video. InProceedings of the 27th ACM international conference on multimedia, pages 2276–2279, 2019. 3

    work page 2019

  9. [9]

    The mapillary traffic sign dataset for detection and classification on a global scale

    Christian Ertler, Jerneja Mislej, Tobias Ollmann, Lorenzo Porzi, Gerhard Neuhold, and Yubin Kuang. The mapillary traffic sign dataset for detection and classification on a global scale. InEuropean conference on computer vision, pages 68–84. Springer, 2020. 1, 2, 3

    work page 2020

  10. [10]

    The pascal visual object classes challenge: A retrospective.Inter- national journal of computer vision, 111(1):98–136, 2015

    Mark Everingham, SM Ali Eslami, Luc Van Gool, Christo- pher KI Williams, John Winn, and Andrew Zisserman. The pascal visual object classes challenge: A retrospective.Inter- national journal of computer vision, 111(1):98–136, 2015. 1

    work page 2015

  11. [11]

    Robust physical-world attacks on deep learning visual classification

    Kevin Eykholt, Ivan Evtimov, Earlence Fernandes, Bo Li, Amir Rahmati, Chaowei Xiao, Atul Prakash, Tadayoshi Kohno, and Dawn Song. Robust physical-world attacks on deep learning visual classification. InProceedings of the IEEE conference on computer vision and pattern recogni- tion, pages 1625–1634, 2018. 1

    work page 2018

  12. [12]

    Traffic sign recognition using local vision transformer

    Ali Farzipour, Omid Nejati Manzari, and Shahriar B Shok- ouhi. Traffic sign recognition using local vision transformer. In2023 13th International Conference on Computer and Knowledge Engineering (ICCKE), pages 191–196. IEEE,

    work page

  13. [13]

    Rich feature hierarchies for accurate object detection and semantic segmentation

    Ross Girshick, Jeff Donahue, Trevor Darrell, and Jitendra Malik. Rich feature hierarchies for accurate object detection and semantic segmentation. InProceedings of the IEEE con- ference on computer vision and pattern recognition, pages 580–587, 2014. 2

    work page 2014

  14. [14]

    Zero-reference deep curve estimation for low-light image enhancement

    Chunle Guo, Chongyi Li, Jichang Guo, Chen Change Loy, Junhui Hou, Sam Kwong, and Runmin Cong. Zero-reference deep curve estimation for low-light image enhancement. In Proceedings of the IEEE/CVF conference on computer vi- sion and pattern recognition, pages 1780–1789, 2020. 2

    work page 2020

  15. [15]

    Traffic Sign Classification Using Deep Inception Based Convolutional Networks

    Mrinal Haloi. Traffic sign classification using deep in- ception based convolutional networks.arXiv preprint arXiv:1511.02992, 2015. 2

    work page internal anchor Pith review Pith/arXiv arXiv 2015

  16. [16]

    Traffic sign recognition based on the nvidia jetson tx1 embedded system using convolutional neural networks

    Yan Han and Erdal Oruklu. Traffic sign recognition based on the nvidia jetson tx1 embedded system using convolutional neural networks. In2017 IEEE 60th International Midwest Symposium on Circuits and Systems (MWSCAS), pages 184–

    work page

  17. [17]

    Deep residual learning for image recognition

    Kaiming He, Xiangyu Zhang, Shaoqing Ren, and Jian Sun. Deep residual learning for image recognition. InProceed- ings of the IEEE conference on computer vision and pattern recognition, pages 770–778, 2016. 7

    work page 2016

  18. [18]

    Cycada: Cycle-consistent adversarial domain adaptation

    Judy Hoffman, Eric Tzeng, Taesung Park, Jun-Yan Zhu, Phillip Isola, Kate Saenko, Alexei Efros, and Trevor Darrell. Cycada: Cycle-consistent adversarial domain adaptation. In International conference on machine learning, pages 1989–

    work page 1989

  19. [19]

    Imagenet classification with deep convolutional neural net- works.Advances in neural information processing systems, 25, 2012

    Alex Krizhevsky, Ilya Sutskever, and Geoffrey E Hinton. Imagenet classification with deep convolutional neural net- works.Advances in neural information processing systems, 25, 2012. 1

    work page 2012

  20. [20]

    Alina Kuznetsova, Hassan Rom, Neil Alldrin, Jasper Ui- jlings, Ivan Krasin, Jordi Pont-Tuset, Shahab Kamali, Stefan Popov, Matteo Malloci, Alexander Kolesnikov, et al. The open images dataset v4: Unified image classification, object detection, and visual relationship detection at scale.Interna- tional journal of computer vision, 128(7):1956–1981, 2020. 1

    work page 1956

  21. [21]

    Using fourier de- scriptors and spatial models for traffic sign recognition

    Fredrik Larsson and Michael Felsberg. Using fourier de- scriptors and spatial models for traffic sign recognition. In Scandinavian conference on image analysis, pages 238–249. Springer, 2011. 2, 7

    work page 2011

  22. [22]

    Perceptual generative adversar- ial networks for small object detection

    Jianan Li, Xiaodan Liang, Yunchao Wei, Tingfa Xu, Jiashi Feng, and Shuicheng Yan. Perceptual generative adversar- ial networks for small object detection. InProceedings of the IEEE conference on computer vision and pattern recog- nition, pages 1222–1230, 2017. 1

    work page 2017

  23. [23]

    Blip: Bootstrapping language-image pre-training for unified vision-language understanding and generation

    Junnan Li, Dongxu Li, Caiming Xiong, and Steven Hoi. Blip: Bootstrapping language-image pre-training for unified vision-language understanding and generation. InInterna- tional conference on machine learning, pages 12888–12900. PMLR, 2022. 7, 8

    work page 2022

  24. [24]

    Microsoft coco: Common objects in context

    Tsung-Yi Lin, Michael Maire, Serge Belongie, James Hays, Pietro Perona, Deva Ramanan, Piotr Doll´ar, and C Lawrence Zitnick. Microsoft coco: Common objects in context. In European conference on computer vision, pages 740–755. Springer, 2014. 1

    work page 2014

  25. [25]

    Feature pyra- mid networks for object detection

    Tsung-Yi Lin, Piotr Doll ´ar, Ross Girshick, Kaiming He, Bharath Hariharan, and Serge Belongie. Feature pyra- mid networks for object detection. InProceedings of the IEEE conference on computer vision and pattern recogni- tion, pages 2117–2125, 2017. 2

    work page 2017

  26. [26]

    Focal loss for dense object detection

    Tsung-Yi Lin, Priya Goyal, Ross Girshick, Kaiming He, and Piotr Doll´ar. Focal loss for dense object detection. InPro- ceedings of the IEEE international conference on computer vision, pages 2980–2988, 2017. 2, 7

    work page 2017

  27. [27]

    Yolo-llts: 9 Real-time low-light traffic sign detection via prior-guided enhancement and multi-branch feature interaction.arXiv preprint arXiv:2503.13883, 2025

    Ziyu Lin, Yunfan Wu, Yuhang Ma, Junzhou Chen, Ronghui Zhang, Jiaming Wu, Guodong Yin, and Liang Lin. Yolo-llts: 9 Real-time low-light traffic sign detection via prior-guided enhancement and multi-branch feature interaction.arXiv preprint arXiv:2503.13883, 2025. 1

    work page arXiv 2025

  28. [28]

    Ssd: Single shot multibox detector

    Wei Liu, Dragomir Anguelov, Dumitru Erhan, Christian Szegedy, Scott Reed, Cheng-Yang Fu, and Alexander C Berg. Ssd: Single shot multibox detector. InEuropean con- ference on computer vision, pages 21–37. Springer, 2016. 1, 2

    work page 2016

  29. [29]

    Image-adaptive yolo for object detec- tion in adverse weather conditions

    Wenyu Liu, Gaofeng Ren, Runsheng Yu, Shi Guo, Jianke Zhu, and Lei Zhang. Image-adaptive yolo for object detec- tion in adverse weather conditions. InProceedings of the AAAI conference on artificial intelligence, pages 1792–1800,

    work page

  30. [30]

    Imagegcn: Multi-relational image graph convolutional networks for dis- ease identification with chest x-rays.IEEE transactions on medical imaging, 41(8):1990–2003, 2022

    Chengsheng Mao, Liang Yao, and Yuan Luo. Imagegcn: Multi-relational image graph convolutional networks for dis- ease identification with chest x-rays.IEEE transactions on medical imaging, 41(8):1990–2003, 2022. 2

    work page 1990

  31. [31]

    Traffic sign recognition—how far are we from the solution? InThe 2013 international joint conference on Neural networks (IJCNN), pages 1–8

    Markus Mathias, Radu Timofte, Rodrigo Benenson, and Luc Van Gool. Traffic sign recognition—how far are we from the solution? InThe 2013 international joint conference on Neural networks (IJCNN), pages 1–8. IEEE, 2013. 2

    work page 2013

  32. [32]

    Subject representation learning from eeg using graph convolutional variational autoencoders

    Aditya Mishra, Ahnaf Mozib Samin, Ali Etemad, and Javad Hashemi. Subject representation learning from eeg using graph convolutional variational autoencoders. InICASSP 2025-2025 IEEE International Conference on Acoustics, Speech and Signal Processing (ICASSP), pages 1–5. IEEE,

    work page 2025

  33. [33]

    Andreas Mogelmose, Mohan Manubhai Trivedi, and Thomas B Moeslund. Vision-based traffic sign detection and analysis for intelligent driver assistance systems: Per- spectives and survey.IEEE transactions on intelligent trans- portation systems, 13(4):1484–1497, 2012. 2

    work page 2012

  34. [34]

    The mapillary vistas dataset for semantic understanding of street scenes

    Gerhard Neuhold, Tobias Ollmann, Samuel Rota Bulo, and Peter Kontschieder. The mapillary vistas dataset for semantic understanding of street scenes. InProceedings of the IEEE international conference on computer vision, pages 4990– 4999, 2017. 2

    work page 2017

  35. [35]

    Day- to-night image synthesis for training nighttime neural isps

    Abhijith Punnappurath, Abdullah Abuolaim, Abdelrahman Abdelhamed, Alex Levinshtein, and Michael S Brown. Day- to-night image synthesis for training nighttime neural isps. InProceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition, pages 10769–10778, 2022. 1

    work page 2022

  36. [36]

    Learning transferable visual models from natural language supervi- sion

    Alec Radford, Jong Wook Kim, Chris Hallacy, Aditya Ramesh, Gabriel Goh, Sandhini Agarwal, Girish Sastry, Amanda Askell, Pamela Mishkin, Jack Clark, et al. Learning transferable visual models from natural language supervi- sion. InInternational conference on machine learning, pages 8748–8763. PmLR, 2021. 2, 7, 8

    work page 2021

  37. [37]

    Yolo9000: better, faster, stronger

    Joseph Redmon and Ali Farhadi. Yolo9000: better, faster, stronger. InProceedings of the IEEE conference on computer vision and pattern recognition, pages 7263–7271, 2017. 2

    work page 2017

  38. [38]

    You only look once: Unified, real-time object de- tection

    Joseph Redmon, Santosh Divvala, Ross Girshick, and Ali Farhadi. You only look once: Unified, real-time object de- tection. InProceedings of the IEEE conference on computer vision and pattern recognition, pages 779–788, 2016. 1, 7

    work page 2016

  39. [39]

    Faster r-cnn: Towards real-time object detection with region proposal networks.Advances in neural information process- ing systems, 28, 2015

    Shaoqing Ren, Kaiming He, Ross Girshick, and Jian Sun. Faster r-cnn: Towards real-time object detection with region proposal networks.Advances in neural information process- ing systems, 28, 2015. 1, 2, 7

    work page 2015

  40. [40]

    Graph convolutional networks for image classification: Comparing approaches for building graphs from images

    J Rodrigues and J Carbonera. Graph convolutional networks for image classification: Comparing approaches for building graphs from images. InProceedings of the 26th International Conference on Enter-prise Information Systems-Volume 1: ICEIS, pages 437–446, 2024. 2

    work page 2024

  41. [41]

    Acdc: The adverse conditions dataset with correspondences for semantic driving scene understanding

    Christos Sakaridis, Dengxin Dai, and Luc Van Gool. Acdc: The adverse conditions dataset with correspondences for semantic driving scene understanding. InProceedings of the IEEE/CVF international conference on computer vision, pages 10765–10775, 2021. 1

    work page 2021

  42. [42]

    Russian traffic sign images dataset.Computer optics, 40(2):294–300, 2016

    Vladislav Igorevich Shakhuro and Anton Sergeevich Konushin. Russian traffic sign images dataset.Computer optics, 40(2):294–300, 2016. 1, 2

    work page 2016

  43. [43]

    The german traffic sign recognition bench- mark: a multi-class classification competition

    Johannes Stallkamp, Marc Schlipsing, Jan Salmen, and Christian Igel. The german traffic sign recognition bench- mark: a multi-class classification competition. InThe 2011 international joint conference on neural networks, pages 1453–1460. IEEE, 2011. 2

    work page 2011

  44. [44]

    Johannes Stallkamp, Marc Schlipsing, Jan Salmen, and Christian Igel. Man vs. computer: Benchmarking machine learning algorithms for traffic sign recognition.Neural net- works, 32:323–332, 2012. 2

    work page 2012

  45. [45]

    Llth-yolov5: a real-time traffic sign detection algorithm for low-light scenes.Automotive Innovation, 7(1): 121–137, 2024

    Xiaoqiang Sun, Kuankuan Liu, Long Chen, Yingfeng Cai, and Hai Wang. Llth-yolov5: a real-time traffic sign detection algorithm for low-light scenes.Automotive Innovation, 7(1): 121–137, 2024. 1, 2

    work page 2024

  46. [46]

    Traffic sign detec- tion and recognition based on convolutional neural network

    Ying Sun, Pingshu Ge, and Dequan Liu. Traffic sign detec- tion and recognition based on convolutional neural network. In2019 Chinese automation congress (CAC), pages 2851–

    work page

  47. [47]

    Training data-efficient image transformers & distillation through at- tention

    Hugo Touvron, Matthieu Cord, Matthijs Douze, Francisco Massa, Alexandre Sablayrolles, and Herv ´e J´egou. Training data-efficient image transformers & distillation through at- tention. InInternational conference on machine learning, pages 10347–10357. PMLR, 2021. 7, 8

    work page 2021

  48. [48]

    Low-light image enhancement framework for improved object detection in fisheye lens datasets

    Dai Quoc Tran, Armstrong Aboah, Yuntae Jeon, Maged Shoman, Minsoo Park, and Seunghee Park. Low-light image enhancement framework for improved object detection in fisheye lens datasets. InProceedings of the IEEE/CVF Con- ference on Computer Vision and Pattern Recognition (CVPR) Workshops, pages 7056–7065, 2024. 2

    work page 2024

  49. [49]

    Cnn2graph: Build- ing graphs for image classification

    Vivek Trivedy and Longin Jan Latecki. Cnn2graph: Build- ing graphs for image classification. InProceedings of the IEEE/CVF Winter Conference on Applications of Computer Vision, pages 1–11, 2023. 2

    work page 2023

  50. [50]

    In- dian traffic sign detection and classification through a unified framework.IEEE Transactions on Intelligent Transportation Systems, 25(10):14866–14875, 2024

    Rishabh Uikey, Haroon R Lone, and Akshay Agarwal. In- dian traffic sign detection and classification through a unified framework.IEEE Transactions on Intelligent Transportation Systems, 25(10):14866–14875, 2024. 3, 7, 8, 6

    work page 2024

  51. [51]

    Learning to navigate for fine-grained clas- sification

    Ze Yang, Tiange Luo, Dong Wang, Zhiqiang Hu, Jun Gao, and Liwei Wang. Learning to navigate for fine-grained clas- sification. InProceedings of the European conference on computer vision (ECCV), pages 420–435, 2018. 2

    work page 2018

  52. [52]

    BD D100K: A diverse driving video database with scalable annotation tooling,

    Fisher Yu, Wenqi Xian, Yingying Chen, Fangchen Liu, Mike Liao, Vashisht Madhavan, Trevor Darrell, et al. Bdd100k: A 10 diverse driving video database with scalable annotation tool- ing.arXiv preprint arXiv:1805.04687, 2(5):6, 2018. 2

    work page arXiv 2018

  53. [53]

    Unpaired image-to-image translation using cycle- consistent adversarial networks

    Jun-Yan Zhu, Taesung Park, Phillip Isola, and Alexei A Efros. Unpaired image-to-image translation using cycle- consistent adversarial networks. InProceedings of the IEEE international conference on computer vision, pages 2223– 2232, 2017. 1

    work page 2017

  54. [54]

    Traffic-sign detection and classifica- tion in the wild

    Zhe Zhu, Dun Liang, Songhai Zhang, Xiaolei Huang, Baoli Li, and Shimin Hu. Traffic-sign detection and classifica- tion in the wild. InProceedings of the IEEE conference on computer vision and pattern recognition, pages 2110–2118,

    work page

  55. [55]

    1, 2 11 Navigating in the Dark: A Multimodal Framework and Dataset for Nighttime Traffic Sign Recognition Supplementary Material

    work page

  56. [56]

    A summary of all figures is given in Table 4

    Indian Nighttime Traffic Sign Dataset This section provides representative examples from our dataset, illustrating the various challenges, supplementary classes, weather conditions, and daytime scenarios detailed in Section 3. A summary of all figures is given in Table 4. Table 4. Summary of all the figures. Figure Description Figure 7 Sensor and weather ...

    work page

  57. [57]

    Figure 12

    The top-row samples, captured using smartphones, generally exhibit higher visual quality compared to the bottom-row images captured with an action camera. Figure 12. Images with multiple signboards in one frame, creating the kind of visual clutter common on Indian roads. Both images have been taken from smartphone. Figure 13. Four distinct signboard desig...

    work page

  58. [58]

    All hyperparameters and training set- tings reported here correspond to those used in our study

    Implementation Details This section provides the implementation details for both components of our pipeline: (i) LENS-Net detector, and (ii) LENS-Net classifier. All hyperparameters and training set- tings reported here correspond to those used in our study. We used a system with a 16 GB NVIDIA GeForce RTX 5080 to train the model. The LENS-Net detector an...

    work page

  59. [59]

    anchor” or “context

    Method We provide the complete architecture of the LENS-Net de- tector in Algorithm 2. For a given image cropI crop from the LENS-Net detector, we pass it through the frozen CLIP Vision Transformer (ViT). We extract the output embed- ding from the CLIP image encoder (E img clip ), yielding an L2-normalized global feature vectorv∈R D. This vector serves as...

    work page

  60. [60]

    We also provide the class-wise performance for Rishabh et al

    Additional Results Table 8 reports the per-class precision, recall, and support across all 41 traffic-sign categories for the LENS-Net clas- sifier, averaged over the five cross-validation folds. We also provide the class-wise performance for Rishabh et al

    work page

  61. [61]

    The results show that the LENS-Net classifier maintains strong perfor- mance even for classes with extremely limited samples

    (the strongest baseline) for comparison. The results show that the LENS-Net classifier maintains strong perfor- mance even for classes with extremely limited samples. For rare classes, such asairport,blow horn,cross road, and steep ascent, which contain very few (relative to dominant classes) samples per fold, the model achieves non-trivial precision or r...

    work page

  62. [62]

    First, it ensures that each image in the dataset is used for both training and testing, thereby reduc- ing bias in the evaluation process

    Frequently Asked Questions Q: Why do you use 5-fold cross-validation? The use of stratified 5-fold cross-validation is crucial for several reasons. First, it ensures that each image in the dataset is used for both training and testing, thereby reduc- ing bias in the evaluation process. Second, by averaging the performance metrics across all folds, we obta...

    work page