Pith. sign in

REVIEW 4 major objections 4 minor 70 references

Adapt, But Don't Forget: Fine-Tuning and Contrastive Routing for Lane Detection under Distribution Shift

T0 review · 4 major / 4 minor · reviewed 2026-08-06 · deepseek-v4-flash

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

Pith's one-line read The paper claims that modular per-distribution branches plus contrastive routing adapt lane detectors to new datasets while preserving source performance and using fewer parameters than full fine-tuning.

desk verdict Solid component-wise fine-tuning study; the routing claim outruns the evidence — Table 3 is a branch-score average, not an end-to-end routed F1. read the letter →

arxiv 2507.18653 v1 pith:QEOO3SB3 submitted 2025-07-22 cs.CV cs.LG

Mohammed Abdul Hafeez Khan , Parth Ganeriwala , Sarah M. Lehman , Siddhartha Bhattacharyya , Amy Alvarez , Natasha Neogi This is my paper
classification cs.CVcs.LG
keywords lanedetectiondistributionshiftcatastrophicforgettingfine-tuningsupervisedcontrastivelearningmodelroutingparameter-efficientadaptation
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

Lane detection models trained on one dataset lose their original skill when fine-tuned on another, even when both are road scenes; the paper calls this catastrophic forgetting and shows it can be severe enough to drop source F1 from 80 to near zero. The proposed remedy is to keep the source model intact and add cloned branches for each new distribution, fine-tuning only the components that matter—the detection head and feature neck for mild shifts, and deeper backbone layers as well for large shifts. At inference, a supervised contrastive classifier decides which distribution an image comes from and routes it to the corresponding branch. The paper claims this reaches near-optimal F1 on all three datasets it tests while using as little as 41% of the parameters of training separate full models, with source performance unchanged.

What carries the argument

The machinery is a modular version of the anchor-based detector CLRerNet, split into backbone B, neck N, and head H, where N and H are cloned and fine-tuned per target while B stays frozen in the light setting, or the last k backbone layers are included in the clone for severe shifts. Inference is driven by a supervised contrastive learning (SCL) classifier: a frozen ResNet encoder with a projection MLP is trained with the SupCon loss so that embeddings of the same distribution cluster tightly, then each distribution is represented by a centroid and an input is assigned to the nearest centroid, selecting the matching branch. The routing point X in 'Routing@X' is the layer at which per-distribution specialization starts; layers before X are shared, layers from X onward are cloned.

What would settle it

Run the full routing pipeline end-to-end on the three test sets and compare the actual routed F1 with the oracle-routed F1 reported in Table 3; if the 99.6–99.9% classifier accuracy does not translate into matching end-to-end scores, the claim of near-optimal routed performance fails. A sharper test is to feed in a held-out fourth distribution that is visually close to one of the three, for example night-time CULane, and check whether the classifier forces it to the nearest centroid and the wrong branch drops F1.

Watch

Extended reading notes

Core claim

The core claim is that adaptation to a new lane-detection distribution and retention of the source distribution need not trade off against each other if the network is reorganized into frozen shared modules plus per-distribution branches. Across three datasets—CULane, CurveLanes, and AssistTaxi—the paper shows that fine-tuning only the neck and head preserves most target accuracy for moderate shifts, while for a severe shift such as airport taxiways from a car-lane base, unfreezing the last two backbone layers plus neck and head is required. With CULane as source, routing at neck+head gives an average F1 of 80.8 with 41% of full fine-tuning's parameters, against 83.6 for full adaptation. With AssistTaxi as source, routing at partial backbone gives 79.2 with 96% of parameters, against 82.3 for full adaptation. Because the source branch is never updated, source F1 is retained by construction. A supervised contrastive learning classifier trained on the three distributions reports 99.6–99.9% accuracy and routes each input to the matching branch.

Load-bearing premise

The system assumes every test image belongs to exactly one of the known training distributions and that the contrastive classifier always sends it to the right branch; a single wrong route into the wrong branch can produce near-zero F1 for extreme shifts.

Editorial extensions

If this is right

  • Source performance is preserved exactly, because the source branch parameters are never modified during adaptation.
  • For moderate shifts such as CULane to CurveLanes or AssistTaxi, neck+head branches match most of the full fine-tuning gain with only about 41% of the parameters (6.8M versus 16.7M on ERFNet).
  • For severe shifts such as AssistTaxi to road lanes, shallow branches fail, with average F1 of 32.5–43.6, while partial-backbone branches reach 79.2 average F1 at 96% of the parameters.
  • The 0.31 ms added by contrastive routing on top of 9.82 ms base inference keeps the system real-time on an RTX 4080.
  • The component-wise observations give a heuristic for choosing the branch point: the larger the distribution change, the deeper the fine-tuned modules must go.

Reading between the lines

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

  • A testable extension is to predict the needed routing depth from an embedding-space distance between source and target, such as the SCL centroid distance, rather than from empirical sweeps; the paper does not make this link.
  • The closed-set assumption is the natural boundary: an unknown distribution will be forced to the nearest known centroid, so adding a rejection rule or an 'unknown' class to the SCL classifier would be needed for open-world deployment.
  • The same modular branch-and-route scheme likely transfers to other anchor- or row-wise lane detectors whose heads encode positional priors, since only the component roles are used, but this remains to be demonstrated.
  • As more target distributions are added, branch count and memory grow linearly; distilling routed predictions back into a single shared network, which the paper only suggests as future work, would trade a small accuracy loss for constant parameter growth.
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

4 major / 4 minor

Summary. The paper proposes a modular branching framework for adapting CLRerNet lane detection models to multiple distributions while avoiding catastrophic forgetting. The authors first train on a source dataset, then create cloned branches for each target and fine-tune only selected components (bias, head, neck+head, partial backbone, full) based on a component-wise analysis across CULane, CurveLanes, and AssistTaxi with DLA-34, ResNet-18, and ERFNet backbones. At inference, a supervised contrastive learning (SCL) classifier estimates the input's distribution and routes it to the corresponding branch. Table 3 reports per-configuration F1 averages and parameter counts, and the paper claims near-optimal F1 with fewer parameters than training separate models.

Significance. The component-wise forgetting analysis is a potentially useful empirical contribution: it covers three backbones and three datasets, reports source and target F1 drops/gains, and gives transparent parameter counts. The SCL routing overhead (0.31 ms) and near-perfect per-dataset classification accuracy are also promising. However, the central claim of the paper—that the inference-time routing framework achieves near-optimal F1 with fewer parameters—is not yet demonstrated, because the joint system is never evaluated and the routing-depth choice is made from the test table. The strengths are real but do not currently support the abstract's headline claim.

major comments (4)
  1. [§4.4, Table 3, Eq. (9)] The central evaluation is missing: Eq. (9) defines the routed output, but the paper never computes an end-to-end F1 on a mixed test set. Table 3's F1Avg is the unweighted arithmetic mean of the three per-distribution branch F1s; because F1 Src is constant by construction and the SCL classifier is evaluated separately in §4.4, F1Avg does not reflect misrouting. With reported SCL accuracy of at most 99.9%, on a mixed test set roughly 0.1–0.4% of inputs would be sent to a branch whose target F1 can be near zero (e.g., an AssistTaxi N+H branch on a CULane image has F1 0.8 in Table 5). The abstract's 'near-optimal F1-scores' is therefore not directly supported until the joint system is evaluated (e.g., a mixed test stream with the confusion matrix and the routed F1).
  2. [Table 3] The highlighted routing configurations are selected based on the test-set F1Avg. For CULane source, Routing@N+H (F1Avg 80.8) is claimed best even though Routing@B(k=2)+N+H and Routing@B+N+H both reach 83.6; for AssistTaxi source, Routing@B(k=2)+N+H (79.2) is chosen over Routing@B+N+H (82.3) and Routing@N+H (43.6). No validation-based rule for choosing the routing depth is given, so the '41% of parameters' headline compares a configuration that was picked after inspecting the test table against a full fine-tuning baseline. Please either select via a validation split or report all routing options without asserting a single best configuration.
  3. [§3.4] The supervised contrastive classifier is central to routing but its training is underspecified: Eq. (5) defines the loss, yet the paper does not state the encoder/MLP architecture, optimizer, learning rate, epochs, batch size, temperature, augmentation, or how the centroids in Eq. (6) are computed. These details are needed to reproduce the claimed 99.6–99.9% accuracy and to judge whether the classifier would transfer to a genuinely mixed test stream rather than being fitted to the same three known splits.
  4. [Table 2 vs. Table 3] Table 2 reports AssistTaxi source F1 = 0.0 for every fine-tuning configuration, including B+N+H, while Table 3 reports F1 Src = 94.3 for the corresponding Routing@B+N+H. If Table 2 describes standard fine-tuning without branch preservation, the text should say so explicitly; if it describes the proposed branching, it contradicts the 'no source forgetting' claim. This apparent inconsistency needs to be resolved.
minor comments (4)
  1. [Eq. (7)] Equation (7) contains a typographical error in the subscript: the argmin is written over 'k in 1,...,K ; |...|' with a semicolon inside the expression; it should be 'argmin_k ||g_psi(enc_phi(x)) - mu_k||_2^2'.
  2. [Table 3 caption] The caption says F1 Src is constant across configurations and is 'denoted with ' ', but the table entries for F1 Src appear blank; please use a visible placeholder or repeat the values to avoid confusion.
  3. [Table 3, Relative column] The 'Relative' column treats Routing@B+N+H as 100%, but the surrounding text says the comparison is against 'fine-tuning fully separate models (treated as baseline)'; clarify whether the baseline is separate full models or the branched full-routing configuration.
  4. [§4.3] No error bars or multiple-seed results are reported; given that some Table 3 differences are only 2–3 F1 points (e.g., 80.8 vs. 83.6), it would be helpful to state whether these gaps are stable across seeds.

Circularity Check

2 steps flagged · score 6.0 of 10

The headline F1 is an unweighted branch-score average rather than the routed output of Eq. (9), and the routing configuration is selected from the same test table used to report near-optimal performance, making the central claim partly reduce to its inputs.

  1. self definitional [Section 3.4, Eq. (9); Section 4.4, Table 3 caption]
    "The final output is therefore: ŷ = f_{θ′_{d̂(x)}}(x). ... Table 3. ... Each block corresponds to a source distribution, with F1 Src constant across configurations (denoted with “ ”). ... F1Avg Params (M) Relative"

    Table 3 computes F1Avg as the unweighted mean of the per-branch F1 columns while keeping the source F1 constant; it never routes a mixed test set through the SCL classifier plus branch. The quantity reported as the framework's F1 is therefore, by construction, just a summary of branch-only test scores, not the F1 of Eq. (9). Any claim that 'the framework achieves near-optimal F1-scores' using this F1Avg assumes perfect routing and ignores the router's measured 99.6–99.9% accuracy and the near-zero cross-branch F1s. The headline result is thus definitionally the branch scores that went in, not an evaluated property of the routing framework.

  2. fitted input called prediction [Section 4.4, Table 3 discussion and Figure 1 caption]
    "When the source-target shift is moderate (CULane →CurveLanes/AssistTaxi), Routing@N+H strikes the best balance, achieving near-optimal performance with only 41% of the parameters. ... Routing@ B(k=2)+N+H yields the best results for AssistTaxi (see Table 3, highlighted in green)."

    The routing point is chosen by scanning the test-set Table 3 for the highest Avg F1, and the same highlighted rows are then used to support the abstract's 'near-optimal F1-scores while using significantly fewer parameters.' No validation split or a priori rule is specified. The 'best' configuration is a post-hoc maximum over the test set, so the 41% and 96% parameter-efficiency figures are properties of the selected test row, not independent predictions. The component-wise analysis and the 'effective fine-tuning strategy' are therefore the same test numbers used twice.

full rationale

The paper is mostly a straightforward empirical study: component-wise fine-tuning is measured on per-dataset test splits, the SCL classifier is trained and evaluated on train/test data, and the branching architecture is not derived from the outcome. There is no load-bearing self-citation: references [10], [11], [12], and [26] are the authors' own datasets or earlier methods, but they are used as evaluation material or related work rather than to justify the central claim. The circularity is concentrated in the reporting of the routing framework. Eq. (9) defines the routed output, but Table 3's F1Avg is an unweighted mean of per-branch F1s with the source F1 constant, so it is not a joint evaluation of the router plus branches. The F1 attributed to the framework is therefore, by construction, the branch F1s that went in, not the output of Eq. (9). Additionally, the choice between Routing@N+H and Routing@B(k=2)+N+H is made by maximizing Avg F1 on the same test table, making the 'best' configuration and the parameter-efficiency percentages test-selected maxima rather than independent predictions. These two issues make the headline claim partially circular, though the underlying branch-adaptation measurements and the SCL classifier itself remain independently meaningful. Overall score 6: one or more central 'predictions' reduce, at least in part, to their inputs by construction.

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

The ledger shows that the central claim depends on two hand-picked free parameters (fine-tuning schedule, routing depth selection) and one unspecified hyperparameter set (SCL). The domain assumptions (labeled targets, discrete known distributions, modular separability) are reasonable for a research setting but are not justified for real deployment.

free parameters (4)
  • Fine-tuning epochs = 3
    Chosen based on observed F1 saturation in Figure 3; the paper states 'we run for three epochs' without a formal criterion.
  • Fine-tuning learning rate = 6e-4
    Fixed for all fine-tuning experiments; no learning rate sweep is reported, so the value is a hand-picked constant.
  • Routing depth per source = N+H for CULane/CurveLanes, B(k=2)+N+H for AssistTaxi
    Selected post hoc as the configuration with the best Avg F1 in Table 3 for each source; this is a fitted categorical choice.
  • SCL model hyperparameters = unspecified
    Temperature, epochs, batch size, optimizer, and augmentation for the supervised contrastive model are not given in the paper, yet the reported 99.6-99.9% accuracy depends on them.
assumptions (4)
  • domain assumption Labeled target data is available for fine-tuning
    Problem statement (Sec. 3.2) defines DT as labeled; in real deployment, target labels may be unavailable or costly.
  • domain assumption Test inputs come from exactly one of the K known distributions
    The routing classifier (Sec. 3.4) assigns inputs to one of K centroids; it does not handle open-set or mixed-distribution inputs.
  • domain assumption Network modules (backbone B, neck N, head H) can be independently cloned and frozen without harming adaptation
    The method relies on modular fine-tuning (Sec. 3.3); the paper does not analyze interactions between modules or shared computation.
  • standard math F1 evaluation uses standard linear assignment with tIoU=0.5 and 30-pixel lane rendering width
    Adopted from CLRerNet evaluation protocol (Sec. 4.2); this is a standard domain convention.

how reviews work

0 comments
Cite this review

Pith. "Pith review of Adapt, But Don't Forget: Fine-Tuning and Contrastive Routing for Lane Detection under Distribution Shift." pith.science (2026). https://pith.science/paper/QEOO3SB3

@misc{pith2026250718653,
  author       = {Pith},
  title        = {Pith review of: Adapt, But Don't Forget: Fine-Tuning and Contrastive Routing for Lane Detection under Distribution Shift},
  year         = {2026},
  howpublished = {\url{https://pith.science/paper/QEOO3SB3}},
  note         = {Machine review of arXiv:2507.18653}
}
read the original abstract

Lane detection models are often evaluated in a closed-world setting, where training and testing occur on the same dataset. We observe that, even within the same domain, cross-dataset distribution shifts can cause severe catastrophic forgetting during fine-tuning. To address this, we first train a base model on a source distribution and then adapt it to each new target distribution by creating separate branches, fine-tuning only selected components while keeping the original source branch fixed. Based on a component-wise analysis, we identify effective fine-tuning strategies for target distributions that enable parameter-efficient adaptation. At inference time, we propose using a supervised contrastive learning model to identify the input distribution and dynamically route it to the corresponding branch. Our framework achieves near-optimal F1-scores while using significantly fewer parameters than training separate models for each distribution.

Figures

Figures reproduced from arXiv: 2507.18653 by the authors.

Figure 1
Figure 1. (a) Overview of CLRerNet. (b.1) and (c.1) illustrate one routing configuration. In Step 1, the model is trained on the source distribution (CULane). In Steps 2 and 3, B is frozen, and the N and H modules are cloned and fine-tuned for each target distribution. In Step 4, a supervised contrastive learning (SCL) model is trained to classify the input distribution. During inference (c.1), the SCL model routes shared B f… view at source ↗
Figure 2
Figure 2. t-SNE visualization of distribution embeddings before and after contrastive learning. [PITH_FULL_IMAGE:figures/full_fig_p005_2.png] view at source ↗
Figure 3
Figure 3. F1-scores plateau after three fine-tuning epochs, with [PITH_FULL_IMAGE:figures/full_fig_p005_3.png] view at source ↗
Figures from the paper (1 more)
Figure 4
Figure 4. Figure 4: Comparing AssistTaxi → CULane. H: Fine-tuning the head adjusts anchor regressors, detecting the center lane but failing on sides due to fixed upstream features. N+H: Adapting the neck improves detection for left and center lanes, but right lane remains undetected due t…

Discussion (0). Sign in to comment.

Reference graph

Works this paper leans on

70 extracted references · 55 canonical work pages

  1. [1]

    Bsnet: Lane detection via draw b-spline curves nearby

    Haoxin Chen, Mengmeng Wang, and Yong Liu. Bsnet: Lane detection via draw b-spline curves nearby. arXiv preprint arXiv:2301.06910, 2023. 1, 2

    arXiv 2023

  2. [2]

    Mmdetection: Open mmlab detection tool- box and benchmark

    Kai Chen, Jiaqi Wang, Jiangmiao Pang, Yuhang Cao, Yu Xiong, Xiaoxiao Li, Shuyang Sun, Wansen Feng, Ziwei Liu, Jiarui Xu, et al. Mmdetection: Open mmlab detection tool- box and benchmark. arXiv preprint arXiv:1906.07155, 2019. 5

    arXiv 1906

  3. [3]

    Deeplab: Semantic image segmentation with deep convolutional nets, atrous convolu- tion, and fully connected crfs

    Liang-Chieh Chen, George Papandreou, Iasonas Kokkinos, Kevin Murphy, and Alan L Yuille. Deeplab: Semantic image segmentation with deep convolutional nets, atrous convolu- tion, and fully connected crfs. IEEE transactions on pattern analysis and machine intelligence, 40(4):834–848, 2017. 1

    2017

  4. [4]

    Rethinking atrous convolution for seman- tic image segmentation

    Liang-Chieh Chen, George Papandreou, Florian Schroff, and Hartwig Adam. Rethinking atrous convolution for seman- tic image segmentation. arXiv preprint arXiv:1706.05587 ,

    arXiv

  5. [5]

    Do- main adaptation for semantic segmentation with maximum squares loss

    Minghao Chen, Hongyang Xue, and Deng Cai. Do- main adaptation for semantic segmentation with maximum squares loss. In Proceedings of the IEEE/CVF international conference on computer vision, pages 2090–2099, 2019. 2

    work page 2019

  6. [6]

    A simple framework for contrastive learning of visual representations

    Ting Chen, Simon Kornblith, Mohammad Norouzi, and Ge- offrey Hinton. A simple framework for contrastive learning of visual representations. In International conference on ma- chine learning, pages 1597–1607. PmLR, 2020. 2

    2020

  7. [7]

    Weakly-supervised domain adaptive semantic segmentation with prototypical contrastive learning

    Anurag Das, Yongqin Xian, Dengxin Dai, and Bernt Schiele. Weakly-supervised domain adaptive semantic segmentation with prototypical contrastive learning. In Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition, pages 15434–15443, 2023. 2

    work page 2023

  8. [8]

    An image is worth 16x16 words: Trans- formers for image recognition at scale

    Alexey Dosovitskiy, Lucas Beyer, Alexander Kolesnikov, Dirk Weissenborn, Xiaohua Zhai, Thomas Unterthiner, Mostafa Dehghani, Matthias Minderer, Georg Heigold, Syl- vain Gelly, et al. An image is worth 16x16 words: Trans- formers for image recognition at scale. arXiv preprint arXiv:2010.11929, 2020. 1

    arXiv 2010

Show all 70 references
  1. [9]

    Few- shot object detection with attention-rpn and multi-relation detector

    Qi Fan, Wei Zhuo, Chi-Keung Tang, and Yu-Wing Tai. Few- shot object detection with attention-rpn and multi-relation detector. In Proceedings of the IEEE/CVF conference on computer vision and pattern recognition, pages 4013–4022,

  2. [10]

    Assisttaxi: A comprehensive dataset for taxiway analysis and autonomous operations

    Parth Ganeriwala, Siddhartha Bhattacharyya, Sean Gun- ther, Brian Kish, Mohammed Abdul Hafeez Khan, Ankur Dhadoti, and Natasha Neogi. Assisttaxi: A comprehensive dataset for taxiway analysis and autonomous operations. In 2023 International Conference on Machine Learning and Ap...

    2023

  3. [11]

    Cross dataset analysis and network architecture repair for autonomous car lane detection

    Parth Ganeriwala, Siddhartha Bhattacharyya, and Raja Muthalagu. Cross dataset analysis and network architecture repair for autonomous car lane detection. In 2023 IEEE In- telligent Vehicles Symposium (IV) , pages 1–6. IEEE, 2023. 1

    2023

  4. [12]

    Runway vs

    Parth Ganeriwala, Amy Alvarez, Abdullah AlQahtani, Sid- dhartha Bhattacharyya, Mohammed Abdul Hafeez Khan, and Natasha Neogi. Runway vs. taxiway: Challenges in au- tomated line identification and notation approaches. In 2025 IEEE International systems Conference (SysCon) , pag...

    2025

  5. [13]

    Bootstrap your own latent-a new approach to self-supervised learning

    Jean-Bastien Grill, Florian Strub, Florent Altch ´e, Corentin Tallec, Pierre Richemond, Elena Buchatskaya, Carl Doersch, Bernardo Avila Pires, Zhaohan Guo, Mohammad Ghesh- laghi Azar, et al. Bootstrap your own latent-a new approach to self-supervised learning. Advances in neur...

    2020

  6. [14]

    Domain adaptation for medical image analysis: a survey

    Hao Guan and Mingxia Liu. Domain adaptation for medical image analysis: a survey. IEEE Transactions on Biomedical Engineering, 69(3):1173–1185, 2021. 1

    2021

  7. [15]

    Laneformer: Object- aware row-column transformers for lane detection

    Jianhua Han, Xiajun Deng, Xinyue Cai, Zhen Yang, Hang Xu, Chunjing Xu, and Xiaodan Liang. Laneformer: Object- aware row-column transformers for lane detection. In Pro- ceedings of the AAAI conference on artificial intelligence , pages 799–807, 2022. 1, 2

    2022

  8. [16]

    Expanding low-density latent regions for open-set object detection

    Jiaming Han, Yuqiang Ren, Jian Ding, Xingjia Pan, Ke Yan, and Gui-Song Xia. Expanding low-density latent regions for open-set object detection. In Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition , pages 9591–9600, 2022. 2

    2022

  9. [17]

    Deep residual learning for image recognition

    Kaiming He, Xiangyu Zhang, Shaoqing Ren, and Jian Sun. Deep residual learning for image recognition. In Proceed- ings of the IEEE conference on computer vision and pattern recognition, pages 770–778, 2016. 1

    2016

  10. [18]

    Mask r-cnn

    Kaiming He, Georgia Gkioxari, Piotr Doll ´ar, and Ross Gir- shick. Mask r-cnn. In Proceedings of the IEEE international conference on computer vision, pages 2961–2969, 2017. 1

    2017

  11. [19]

    Momentum contrast for unsupervised visual rep- resentation learning

    Kaiming He, Haoqi Fan, Yuxin Wu, Saining Xie, and Ross Girshick. Momentum contrast for unsupervised visual rep- resentation learning. In Proceedings of the IEEE/CVF con- ference on computer vision and pattern recognition , pages 9729–9738, 2020. 2

    2020

  12. [20]

    Clrernet: improving con- fidence of lane detection with laneiou

    Hiroto Honda and Yusuke Uchida. Clrernet: improving con- fidence of lane detection with laneiou. In Proceedings of the IEEE/CVF winter conference on applications of computer vision, pages 1176–1185, 2024. 1, 2, 4, 5

    2024

  13. [21]

    Learning lightweight lane detection cnns by self at- tention distillation

    Yuenan Hou, Zheng Ma, Chunxiao Liu, and Chen Change Loy. Learning lightweight lane detection cnns by self at- tention distillation. In Proceedings of the IEEE/CVF inter- national conference on computer vision , pages 1013–1021,

  14. [22]

    Mic: Masked image consistency for context- enhanced domain adaptation

    Lukas Hoyer, Dengxin Dai, Haoran Wang, and Luc Van Gool. Mic: Masked image consistency for context- enhanced domain adaptation. In Proceedings of the IEEE/CVF conference on computer vision and pattern recognition, pages 11721–11732, 2023. 2

    2023

  15. [23]

    Sim-to- real domain adaptation for lane detection and classification in autonomous driving

    Chuqing Hu, Sinclair Hudson, Martin Ethier, Mohammad Al-Sharman, Derek Rayside, and William Melek. Sim-to- real domain adaptation for lane detection and classification in autonomous driving. In 2022 IEEE Intelligent Vehicles Symposium (IV), pages 457–463. IEEE, 2022. 2

    2022

  16. [24]

    Measuring catastrophic for- getting in neural networks

    Ronald Kemker, Marc McClure, Angelina Abitino, Tyler Hayes, and Christopher Kanan. Measuring catastrophic for- getting in neural networks. In Proceedings of the AAAI con- ference on artificial intelligence, 2018. 2

    2018

  17. [25]

    Clas- sification of microstructure images of metals using transfer learning

    Mohammed Abdul Hafeez Khan, Hrishikesh Sabnis, J An- gel Arul Jothi, J Kanishkha, and AM Deva Prasad. Clas- sification of microstructure images of metals using transfer learning. In International Conference on Modelling and De- velopment of Intelligent Systems , pages 136–147....

  18. [26]

    Alina: Advanced line identification and notation algorithm

    Mohammed Abdul Hafeez Khan, Parth Ganeriwala, Sid- dhartha Bhattacharyya, Natasha Neogi, and Raja Muthalagu. Alina: Advanced line identification and notation algorithm. In Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition, pages 7293–7302, 2024. 1

    2024

  19. [27]

    Detection of cavities from oral images using convolutional neural networks

    MA Hafeez Khan, Prasad S Giri, and J Angel Arul Jothi. Detection of cavities from oral images using convolutional neural networks. In 2022 International Conference on Elec- trical, Computer and Energy Technologies (ICECET), pages 1–6. IEEE, 2022. 1

    2022

  20. [28]

    Supervised contrastive learning

    Prannay Khosla, Piotr Teterwak, Chen Wang, Aaron Sarna, Yonglong Tian, Phillip Isola, Aaron Maschinot, Ce Liu, and Dilip Krishnan. Supervised contrastive learning. Advances in neural information processing systems, 33:18661–18673,

  21. [29]

    Overcoming catastrophic forgetting in neu- ral networks

    James Kirkpatrick, Razvan Pascanu, Neil Rabinowitz, Joel Veness, Guillaume Desjardins, Andrei A Rusu, Kieran Milan, John Quan, Tiago Ramalho, Agnieszka Grabska- Barwinska, et al. Overcoming catastrophic forgetting in neu- ral networks. Proceedings of the national academy of sc...

    2017

  22. [30]

    Key points esti- mation and point instance segmentation approach for lane detection

    Yeongmin Ko, Younkwan Lee, Shoaib Azam, Farzeen Mu- nir, Moongu Jeon, and Witold Pedrycz. Key points esti- mation and point instance segmentation approach for lane detection. IEEE Transactions on Intelligent Transportation Systems, 23(7):8949–8958, 2021. 1, 2

    2021

  23. [31]

    Vpgnet: Vanishing point guided network for lane and road marking detection and recognition

    Seokju Lee, Junsik Kim, Jae Shin Yoon, Seunghak Shin, Oleksandr Bailo, Namil Kim, Tae-Hee Lee, Hyun Seok Hong, Seung-Hoon Han, and In So Kweon. Vpgnet: Vanishing point guided network for lane and road marking detection and recognition. In Proceedings of the IEEE inter- nationa...

    1947

  24. [32]

    Multi-level domain adaptation for lane detection

    Chenguang Li, Boheng Zhang, Jia Shi, and Guangliang Cheng. Multi-level domain adaptation for lane detection. In Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition, pages 4380–4389, 2022. 2

    2022

  25. [33]

    Constructing self-motivated pyramid curriculums for cross- domain semantic segmentation: A non-adversarial approach

    Qing Lian, Fengmao Lv, Lixin Duan, and Boqing Gong. Constructing self-motivated pyramid curriculums for cross- domain semantic segmentation: A non-adversarial approach. In Proceedings of the IEEE/CVF International Conference on Computer Vision, pages 6758–6767, 2019. 2

    2019

  26. [34]

    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. In Proceedings of the IEEE conference on computer vision and pattern recogni- tion, pages 2117–2125, 2017. 3

    2017

  27. [35]

    Cond- lanenet: a top-to-down lane detection framework based on conditional convolution

    Lizhe Liu, Xiaohao Chen, Siyu Zhu, and Ping Tan. Cond- lanenet: a top-to-down lane detection framework based on conditional convolution. In Proceedings of the IEEE/CVF international conference on computer vision , pages 3773– 3782, 2021. 1, 2, 5

    2021

  28. [36]

    End- to-end lane shape prediction with transformers

    Ruijin Liu, Zejian Yuan, Tie Liu, and Zhiliang Xiong. End- to-end lane shape prediction with transformers. In Proceed- ings of the IEEE/CVF winter conference on applications of computer vision, pages 3694–3702, 2021. 1, 2

    2021

  29. [37]

    Fully convolutional networks for semantic segmentation

    Jonathan Long, Evan Shelhamer, and Trevor Darrell. Fully convolutional networks for semantic segmentation. In Pro- ceedings of the IEEE conference on computer vision and pat- tern recognition, pages 3431–3440, 2015. 1

    2015

  30. [38]

    Gradient episodic memory for continual learning

    David Lopez-Paz and Marc’Aurelio Ranzato. Gradient episodic memory for continual learning. Advances in neu- ral information processing systems, 30, 2017. 2

    2017

  31. [39]

    Decoupled weight decay regularization

    Ilya Loshchilov and Frank Hutter. Decoupled weight decay regularization. arXiv preprint arXiv:1711.05101, 2017. 1, 2

    2017 arXiv

  32. [40]

    Weakly-supervised cross- domain road scene segmentation via multi-level curriculum adaptation

    Fengmao Lv, Guosheng Lin, Peng Liu, Guowu Yang, Sinno Jialin Pan, and Lixin Duan. Weakly-supervised cross- domain road scene segmentation via multi-level curriculum adaptation. IEEE Transactions on Circuits and Systems for Video Technology, 31(9):3493–3503, 2020. 2

    2020

  33. [41]

    Understanding the role of train- ing regimes in continual learning

    Seyed Iman Mirzadeh, Mehrdad Farajtabar, Razvan Pascanu, and Hassan Ghasemzadeh. Understanding the role of train- ing regimes in continual learning. Advances in Neural Infor- mation Processing Systems, 33:7308–7320, 2020. 2

    2020

  34. [42]

    Lane detection technique based on perspective transformation and histogram analysis for self-driving cars

    Raja Muthalagu, Anudeepsekhar Bolimera, and V Kalaichelvi. Lane detection technique based on perspective transformation and histogram analysis for self-driving cars. Computers & Electrical Engineering, 85:106653, 2020. 2

    2020

  35. [43]

    Towards end-to-end lane detection: an instance segmentation approach

    Davy Neven, Bert De Brabandere, Stamatios Georgoulis, Marc Proesmans, and Luc Van Gool. Towards end-to-end lane detection: an instance segmentation approach. In 2018 IEEE intelligent vehicles symposium (IV) , pages 286–291. IEEE, 2018. 1, 2

    2018

  36. [44]

    Dissecting catastrophic forgetting in continual learning by deep visualization

    Giang Nguyen, Shuan Chen, Thao Do, Tae Joon Jun, Ho-Jin Choi, and Daeyoung Kim. Dissecting catastrophic forgetting in continual learning by deep visualization. arXiv preprint arXiv:2001.01578, 2020. 2

    2001 arXiv

  37. [45]

    Lanecor- rect: Self-supervised lane detection

    Ming Nie, Xinyue Cai, Hang Xu, and Li Zhang. Lanecor- rect: Self-supervised lane detection. International Journal of Computer Vision, pages 1–15, 2025. 2

    2025

  38. [46]

    Spatial as deep: Spatial cnn for traffic scene understanding

    Xingang Pan, Jianping Shi, Ping Luo, Xiaogang Wang, and Xiaoou Tang. Spatial as deep: Spatial cnn for traffic scene understanding. In Proceedings of the AAAI conference on artificial intelligence, 2018. 1, 2, 4

    2018

  39. [47]

    Ultra fast structure- aware deep lane detection

    Zequn Qin, Huanyu Wang, and Xi Li. Ultra fast structure- aware deep lane detection. In Computer Vision–ECCV 2020: 16th European Conference, Glasgow, UK, August 23–28, 2020, Proceedings, Part XXIV 16, pages 276–291. Springer,

    2020

  40. [48]

    Ultra fast deep lane detection with hybrid anchor driven ordinal classification

    Zequn Qin, Pengyi Zhang, and Xi Li. Ultra fast deep lane detection with hybrid anchor driven ordinal classification. IEEE transactions on pattern analysis and machine intelli- gence, 46(5):2555–2568, 2022. 1, 2

    2022

  41. [49]

    Focus on local: Detecting lane marker from bottom up via key point

    Zhan Qu, Huan Jin, Yang Zhou, Zhen Yang, and Wei Zhang. Focus on local: Detecting lane marker from bottom up via key point. In Proceedings of the IEEE/CVF conference on computer vision and pattern recognition , pages 14122– 14130, 2021. 2

    2021

  42. [50]

    Anatomy of catastrophic forgetting: Hidden representations and task semantics

    Vinay V Ramasesh, Ethan Dyer, and Maithra Raghu. Anatomy of catastrophic forgetting: Hidden representations and task semantics. arXiv preprint arXiv:2007.07400, 2020. 2

    2007 arXiv

  43. [51]

    Erfnet: Efficient residual factorized convnet for real-time semantic segmentation

    Eduardo Romera, Jos ´e M Alvarez, Luis M Bergasa, and Roberto Arroyo. Erfnet: Efficient residual factorized convnet for real-time semantic segmentation. IEEE Transactions on Intelligent Transportation Systems, 19(1):263–272, 2017. 8

    2017

  44. [52]

    U- net: Convolutional networks for biomedical image segmen- tation

    Olaf Ronneberger, Philipp Fischer, and Thomas Brox. U- net: Convolutional networks for biomedical image segmen- tation. In Medical image computing and computer-assisted intervention–MICCAI 2015: 18th international conference, Munich, Germany, October 5-9, 2015, proceedings, pa...

    2015

  45. [53]

    Keep your eyes on the lane: Real-time attention-guided lane detection

    Lucas Tabelini, Rodrigo Berriel, Thiago M Paixao, Claudine Badue, Alberto F De Souza, and Thiago Oliveira-Santos. Keep your eyes on the lane: Real-time attention-guided lane detection. In Proceedings of the IEEE/CVF conference on computer vision and pattern recognition , pages...

  46. [54]

    Polylanenet: Lane estimation via deep polynomial regres- sion

    Lucas Tabelini, Rodrigo Berriel, Thiago M Paixao, Claudine Badue, Alberto F De Souza, and Thiago Oliveira-Santos. Polylanenet: Lane estimation via deep polynomial regres- sion. In 2020 25th international conference on pattern recog- nition (ICPR), pages 6150–6156. IEEE, 2021. 1, 2

    2020

  47. [55]

    Adversarial discriminative domain adaptation

    Eric Tzeng, Judy Hoffman, Kate Saenko, and Trevor Darrell. Adversarial discriminative domain adaptation. In Proceed- ings of the IEEE conference on computer vision and pattern recognition, pages 7167–7176, 2017. 2

    2017

  48. [56]

    A keypoint-based global association network for lane detection

    Jinsheng Wang, Yinchao Ma, Shaofei Huang, Tianrui Hui, Fei Wang, Chen Qian, and Tianzhu Zhang. A keypoint-based global association network for lane detection. In Proceed- ings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition, pages 1392–1401, 2022. 2

    2022

  49. [57]

    Exploring cross-image pixel contrast for semantic segmentation

    Wenguan Wang, Tianfei Zhou, Fisher Yu, Jifeng Dai, En- der Konukoglu, and Luc Van Gool. Exploring cross-image pixel contrast for semantic segmentation. In Proceedings of the IEEE/CVF international conference on computer vision, pages 7303–7313, 2021. 2

    2021

  50. [58]

    Curvelane-nas: Unifying lane- sensitive architecture search and adaptive point blending

    Hang Xu, Shaoju Wang, Xinyue Cai, Wei Zhang, Xiaodan Liang, and Zhenguo Li. Curvelane-nas: Unifying lane- sensitive architecture search and adaptive point blending. In Computer Vision–ECCV 2020: 16th European Conference, Glasgow, UK, August 23–28, 2020, Proceedings, Part XV 16...

    2020

  51. [59]

    Lane detection in autonomous vehicles: A systematic review

    Noor Jannah Zakaria, Mohd Ibrahim Shapiai, Rasli Abd Ghani, Mohd Najib Mohd Yassin, Mohd Zamri Ibrahim, and Nurbaiti Wahid. Lane detection in autonomous vehicles: A systematic review. IEEE access, 11:3729–3765, 2023. 1

    2023

  52. [60]

    Recent advances in transfer learning for cross-dataset visual recognition: A problem-oriented perspective

    Jing Zhang, Wanqing Li, Philip Ogunbona, and Dong Xu. Recent advances in transfer learning for cross-dataset visual recognition: A problem-oriented perspective. ACM Comput- ing Surveys (CSUR), 52(1):1–38, 2019. 1

    2019

  53. [61]

    Resa: Recurrent feature-shift ag- gregator for lane detection

    Tu Zheng, Hao Fang, Yi Zhang, Wenjian Tang, Zheng Yang, Haifeng Liu, and Deng Cai. Resa: Recurrent feature-shift ag- gregator for lane detection. In Proceedings of the AAAI con- ference on artificial intelligence, pages 3547–3554, 2021. 1, 2

    2021

  54. [62]

    Clrnet: Cross layer re- finement network for lane detection

    Tu Zheng, Yifei Huang, Yang Liu, Wenjian Tang, Zheng Yang, Deng Cai, and Xiaofei He. Clrnet: Cross layer re- finement network for lane detection. In Proceedings of the IEEE/CVF conference on computer vision and pattern recognition, pages 898–907, 2022. 1, 2

    2022

  55. [63]

    Clrnetv2: A faster and stronger lane detector

    Tu Zheng, Yifei Huang, Yang Liu, Binbin Lin, Zheng Yang, Deng Cai, and Xiaofei He. Clrnetv2: A faster and stronger lane detector. IEEE Transactions on Pattern Analysis and Machine Intelligence, 2025. 1, 2

    2025

  56. [64]

    To- wards weakly-supervised domain adaptation for lane detec- tion

    Jingxing Zhou, Chongzhe Zhang, and J ¨urgen Beyerer. To- wards weakly-supervised domain adaptation for lane detec- tion. In Proceedings of the IEEE/CVF Conference on Com- puter Vision and Pattern Recognition , pages 3553–3563,

  57. [65]

    Unsupervised domain adaptive lane detection via contextual contrast and aggregation

    Kunyang Zhou, Yunjian Feng, and Jun Li. Unsupervised domain adaptive lane detection via contextual contrast and aggregation. arXiv preprint arXiv:2407.13328, 2024. 2

    2024 arXiv

  58. [66]

    Balanced contrastive learn- ing for long-tailed visual recognition

    Jianggang Zhu, Zheng Wang, Jingjing Chen, Yi-Ping Phoebe Chen, and Yu-Gang Jiang. Balanced contrastive learn- ing for long-tailed visual recognition. In Proceedings of the IEEE/CVF conference on computer vision and pattern recognition, pages 6908–6917, 2022. 2

    2022

  59. [67]

    Contrastive learning for lane detection via cross-similarity

    Ali Zoljodi, Sadegh Abadijou, Mina Alibeigi, and Masoud Daneshtalab. Contrastive learning for lane detection via cross-similarity. Pattern Recognition Letters, 185:175–183,

  60. [69]

    • Bias: θ′ c = {bias(θi)}, where i ∈ {H} or {N, H}

    Rationale for Fine-Tuning Configurations Here, we provide the motivations for each configuration θ′ c ⊆ θ, outlining the intuition behind the choice of train- able components and their impact under distribution shift. • Bias: θ′ c = {bias(θi)}, where i ∈ {H} or {N, H}. Updatin...

  61. [70]

    Each table reports both source (CU- Lane) and target performance after selectively fine-tuning model components

    Additional Results We present detailed precision, recall, and F1-score break- downs for all fine-tuning configurations in Tables 4 and 5, corresponding to CurveLanes and AssistTaxi target distri- butions, respectively. Each table reports both source (CU- Lane) and target perfo...

  62. [2024]

    2 Adapt, But Don’t Forget: Fine-Tuning and Contrastive Routing for Lane Detection under Distribution Shift Supplementary Material

Pith tools

Reviewed August 6, 2026 · model on record in the stance chip above.