Pith. sign in

REVIEW 5 major objections 8 minor 51 references

Towards Robust Unsupervised Attention Prediction in Autonomous Driving

T0 review · 5 major / 8 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 Unsupervised driver attention prediction, built from refined pseudo-labels of natural-scene saliency models plus robust augmentation, matches or beats fully supervised methods on three benchmarks and is more robust to corruption and…

desk verdict A useful robustness benchmark and a plausible unsupervised attention model, but the headline robustness claim only holds on one subset and the pseudo-label targets are never validated against human gaze. read the letter →

arxiv 2501.15045 v2 pith:57ZK2NFO submitted 2025-01-25 cs.CV cs.AI

Mengshi Qi , Xiaoyang Bi , Pengfei Zhu , Huadong Ma This is my paper · ORCID
classification cs.CVcs.AI
keywords driverattentionpredictionunsupervisedlearningpseudo-labelssaliencycorruptionrobustnesscentralbiasdataaugmentationautonomousdriving
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

Driver attention prediction normally requires expensive human gaze labels collected in vehicles or laboratories. This paper argues that those labels can be replaced entirely by pseudo-labels generated from off-the-shelf saliency models trained on natural images, provided the pseudo-labels are refined in three ways: mining uncertainty across several sources, embedding traffic-object knowledge from an instance segmenter, and training with a robust augmentation called RoboMixup. On BDD-A, DR(eye)VE, and DADA-2000, the resulting unsupervised model matches or beats fully supervised baselines, and on the new DriverAttention-C corruption benchmark it reduces relative corruption degradation by 58.8% and 52.8% while improving central-bias robustness by 12.4% and 11.4% in KLD and CC. If correct, this removes a major data bottleneck for self-driving attention and makes the predictor more reliable under adverse conditions.

What carries the argument

The load-bearing mechanism is the three-stage pseudo-label refinement pipeline. The Uncertainty Mining Branch (UMB) applies non-local self-attention across multi-scale encoder features and N pseudo-label maps to emit per-pixel uncertainty maps, and couples the predicted attention map to each pseudo-label through an uncertainty-weighted KL-divergence loss, where the predicted log variance weighs each pseudo-label's contribution. The Knowledge Embedding Block (KEB) multiplies each pseudo-label by a binary traffic-object mask produced by a frozen Mask R-CNN, with object categories selected automatically by picking frequent COCO classes whose mean pseudo-label attention is below a threshold, on the reasoning that natural-scene saliency models under-represent traffic objects. RoboMixup replaces the global scalar in Mixup with a pixel-level attention ratio between two samples, selects hard samples per batch by KL divergence to the average attention map, and combines random cropping with RegMixup's auxiliary-loss formulation to counter central bias.

What would settle it

Run the method on DADA-2000 crash frames where the recorded driver fixation lands on an object that is neither salient in natural-scene models nor a member of the boosted COCO categories, such as a fallen cyclist, an animal on the roadway, or scattered debris; if the predicted attention map shows no peak at the human fixation on such frames and per-frame KLD and CC drop below the supervised baselines for those frames, the pseudo-label adequacy assumption fails. A direct quantitative check: compute the mean pseudo-label attention over ground-truth fixation pixels; the claim predicts that the boosted categories account for most of this mass, so measuring the residual mass on unboosted categories across the three datasets would settle the adequacy question.

Watch

Extended reading notes

Core claim

The central claim is that the supervision signal for driver attention need not come from human fixations at all. The paper shows that pseudo-labels produced by two natural-scene saliency models, ML-Net and UNISAL, can be transformed into training targets that rival ground-truth attention maps by combining three mechanisms: an Uncertainty Mining Branch that models disagreements among pseudo-labels as per-pixel uncertainty and optimizes a KL-divergence-based uncertainty loss, a Knowledge Embedding Block that boosts traffic-relevant object classes using a frozen Mask R-CNN segmenter, and RoboMixup, which blends images with attention-weighted pixel mixing, dynamic hard-sample selection, and random-crop regularization. The reported outcome is that the unsupervised model achieves KLD of 1.099, 1.901, and 1.677 and CC of 0.640, 0.510, and 0.488 on BDD-A, DR(eye)VE, and DADA-2000 respectively, comparable to or better than supervised state-of-the-art methods, and that on the six-corruption DriverAttention-C benchmark the RoboMixup variant lowers relative corruption degradation to 0.755 and 0.846 in KLD and CC on BDD-A-C, the best among all compared methods.

Load-bearing premise

The load-bearing premise is that the enhanced pseudo-labels, natural-scene saliency outputs boosted on a mined set of common traffic-object categories, approximate what real drivers actually look at in safety-critical scenes; if those saliency models miss crash-relevant objects outside the boosted categories, the unsupervised training signal is blind to the very things that matter.

Editorial extensions

If this is right

  • If the claim is correct, self-driving attention prediction no longer requires collecting human gaze data, which is expensive and prone to inconsistencies such as averaging eight drivers or relying on in-lab simulation.
  • The robustness numbers imply that attention prediction can tolerate common camera corruptions such as noise, blur, JPEG compression, fog, and snow without being trained on corrupted images, improving safety margins under adverse weather.
  • The central-bias results imply the model keeps attending to off-center objects such as pedestrians, cyclists, and fallen persons that are exactly the ones in accident scenarios, rather than collapsing to a road-center prior.
  • The DriverAttention-C benchmark provides a standardized way to measure robustness of attention predictors across noise, blur, digital, and weather corruptions.
  • The exploratory explainable-decision experiment suggests that predicted attention maps can be plugged into downstream driving-decision models to yield interpretable outputs.

Reading between the lines

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

  • The recipe of multi-source pseudo-labels plus uncertainty mining, knowledge embedding, and robust mixup is domain-agnostic and could be applied to other sparse-label prediction tasks such as pedestrian intent or hazard detection, where gaze labels are equally scarce.
  • The knowledge-mining rule assumes that natural-scene saliency models uniformly under-respond to traffic objects; a natural test is whether boosting categories with low mean attention helps only because of the specific class distribution of MS-COCO, and whether the rule needs re-mining for traffic cultures with different object frequencies.
  • Because the robustness gain from dynamic augmentation, which selects samples with high KL divergence to the average map, is not disentangled from soft-attention mixing, an ablation of the selection rule alone would reveal which part of RoboMixup drives the 58.8% relative degradation reduction.
  • A reader should not expect the same unsupervised gains on novel crash-relevant objects such as debris or animals that are absent from both the saliency models' training data and the MS-COCO taxonomy; for such objects the pseudo-label signal may be missing entirely.
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

5 major / 8 minor

Summary. The paper proposes an unsupervised driver-attention prediction framework that trains only on pseudo-labels produced by natural-scene saliency models, without using human gaze labels from driving datasets. Three components are introduced: an Uncertainty Mining Branch (UMB) that fuses multiple pseudo-label sources with multi-scale features, a Knowledge Embedding Block (KEB) that uses Mask R-CNN to boost attention on traffic-relevant categories mined from pseudo-label statistics, and a RoboMixup data augmentation that combines soft-attention Mixup, a dynamic augmentation strategy, and random-crop regularization. The authors also introduce a corruption benchmark, DriverAttention-C, generated from BDD-A, DR(eye)VE, and DADA-2000, and report that the method matches or surpasses fully supervised state-of-the-art on clean data and improves robustness to corruption and central bias. The central claim is that driver attention can be predicted without traffic gaze labels while being more robust than several supervised baselines.

Significance. If the central claim holds, the paper would make a useful contribution: it relaxes the annotation bottleneck for driver-attention prediction and provides a new corruption benchmark (DriverAttention-C) with 115k frames, alongside an augmentation method (RoboMixup) that reduces central bias. The evaluation is extensive, covering three public datasets, a new benchmark, multiple corruption types, and several ablations. The availability of code and data is a strength. However, the significance is conditional on the adequacy of the pseudo-label learning signal, which is never directly validated against human gaze, and on a more careful reporting of the robustness claims, which are currently overstated relative to the full benchmark results.

major comments (5)
  1. [§3.1-§3.2, Eq. (5)] The unsupervised learning signal is never validated against human gaze. The model trains exclusively on KEB-enhanced pseudo-labels from natural-scene saliency models, and Eq. (5) suppresses all non-mined regions by a factor α=0.3. Table 7 shows that the natural-to-driving domain gap is large (SALICON-trained APB yields KLD 2.109 / CC 0.287 on BDD-A versus 1.036 / 0.657 for BDD-A-trained APB). No experiment reports KLD/CC between the KEB pseudo-labels and the human gaze ground truth on held-out frames. Without such a check, the reported clean and robustness results cannot distinguish learning true driver attention from learning a shifted saliency prior. Please report pseudo-label-to-gaze agreement on each dataset, or an upper-bound ablation that replaces pseudo-labels with ground-truth labels.
  2. [Abstract; Tables 3 and 4] The claimed reductions of 58.8% and 52.8% in relative corruption degradation are specific to BDD-A-C and are computed relative to the UAP baseline (Table 4: Relative mCD 0.412 KLD / 0.472 CC). On DR(eye)VE-C and DADA-2000-C, RUAP does not achieve the best Relative mCD; e.g., DADA-2000-C KLD: DADA 0.044 vs RUAP 1.137, and DR(eye)VE-C KLD: Multi-Branch 0.505 vs RUAP 0.652. The abstract and Section 5.3 should state the dataset and baseline for these headline numbers. The text's explanation that low Relative mCD by other methods reflects 'stable but poor performance' is post hoc; a formal criterion (e.g., Pareto comparison over clean KLD and degradation) is needed before discounting those competitors.
  3. [§4.1-§4.2] The DriverAttention-C construction is under-specified and the reported counts are inconsistent. Section 4.1 reports a total of 115,332 frames and per-subset test counts 6,817 + 7,497 + 4,908 = 19,222, while Section 4.2 ends with 'filter out images with artificial stripes caused by inherent issues of the Cycle-GAN, resulting in 38,444 images.' Please clarify what 38,444 refers to, how many images were removed per corruption type, and whether filtering occurred before or after the train/test split. Because the filtering criterion is based on visual artifacts, it may introduce selection bias; the criterion and its effect on benchmark difficulty should be documented.
  4. [§3.3-§3.4, Eqs. (8)-(10)] The uncertainty loss uses a scalar u_n defined as the spatial mean of the pixel-level uncertainty map U_n, yet the UMB is motivated as producing pixel-level uncertainty maps. This scalarization discards spatial information and conflicts with the design narrative. Either adopt per-pixel uncertainty in the loss (e.g., u_i) or justify the scalar aggregation. Additionally, the '≈' step in Eq. (9) needs a derivation; as written it is not evident that the approximation from the negative log-likelihood to LCE/u_n^2 + log(u_n) holds for arbitrary S and Y_n.
  5. [§5.3, Tables 1-3] No error bars or significance tests are reported for any fully-supervised baseline, while Ours is shown with ±. Because the claim of equivalence or superiority relies on KLD differences as small as 0.03-0.1, the comparison is not fully supported without variance estimates for the baselines. Please report variance over at least three runs of the main baselines, or a paired significance test, for the central clean-benchmark and robustness tables.
minor comments (8)
  1. [Table 7] Typo: 'SALION' should be 'SALICON'.
  2. [§5.3] Typo: 'BDDA-C' appears in the paragraph after Table 3; it should be 'BDD-A-C'.
  3. [§4.1] Typo: 'DriverAttetnion-C' should be 'DriverAttention-C'.
  4. [§3.2, Eq. (2)] The notation 'µP · Mij / Mij' is confusing; it should be written as the mean of µP over the pixels of mask Mij.
  5. [§5.1] The text says ML-Net is selected as the baseline for mCD, but Table 4 uses UAP as the reference. Please state explicitly that different references are used in Tables 3 and 4.
  6. [Abstract] The phrase 'three public datasets and the proposed robustness benchmark' counts four items; consider 'three public datasets and the proposed corruption benchmark'.
  7. [Table 2] It is surprising that training on BDD-A pseudo-labels yields better DADA-2000 test KLD (1.677) than training on DADA-2000 pseudo-labels (1.711). A brief explanation of this cross-dataset result would help the reader assess the pseudo-label quality.
  8. [§3.2] The sentence 'We focus on the most frequent categories, as the remaining categories are likely to be misclassified due to their small proportion' is ambiguous; it suggests rare categories are noisy, but the mining then selects low-attention frequent categories. Please clarify.

Circularity Check

1 steps flagged · score 2.0 of 10

Minor self-referential pseudo-label enhancement in KEB; central benchmark evaluation is external and not circular.

  1. self definitional [Section 3.2 (Knowledge Embedding Block), Eqs. (1)-(5)]
    "Pretrained attention prediction models for natural scenes lack the recognition capabilities for traffic-specific objects (e.g., stop signs), leading to an attention distribution bias ... We leverage this bias for knowledge mining by calculating the mean attention for all pseudo-label instances ... As a result, we mine the following categories as prior knowledge ... The process of generating the segmentation map with prior knowledge can be formally expressed as ... Knowledge Embedding Technique ... ˆYn = Yn · ( ˆM + α)"

    The 'prior knowledge' set is defined by Eq. (3) as the categories whose mean pseudo-label attention V_ci is below η times the average of the first k categories. Eq. (5) then multiplies those same pseudo-labels by (1+α) inside the mined mask and by α outside it. Thus the enhancement is a deterministic function of the pseudo-labels themselves: the categories are selected because they are weak in Yn, and Yn is boosted exactly there. The post-embedding increase of attention in the mined categories is true by construction, not by evidence about driver gaze. This does not, however, force the benchmark KLD/CC numbers, since those are computed against human gaze maps that never enter the mining loop; the central claim therefore is not reduced to this loop.

full rationale

The paper's central evaluation is self-contained against external human-gaze benchmarks. The attention prediction branch is trained on pseudo-labels produced by external natural-scene saliency models (ML-Net, UNISAL) and refined with Mask R-CNN / MS-COCO masks, then scored with KLD and CC against official ground-truth gaze maps of BDD-A, DR(eye)VE, DADA-2000 and their corrupted DriverAttention-C variants. None of these ground truths are used to define the training targets, so the reported superiority is not a construction of the method's own outputs. The one self-referential element is the KEB: Eq. (3) mines low-attention categories from the pseudo-labels and Eq. (5) boosts them, which is an internal bootstrapping loop rather than an independent external source of 'driving knowledge.' Because the ablation and final numbers are measured on human gaze, this loop is not load-bearing for the central claim; it is a minor self-referential strengthening. The citation to the authors' ICCV paper [19] is transparent and not used to justify the method's validity. RoboMixup is compared fairly against external augmentation baselines. Overall circularity is minor, so the score is 2 rather than 0.

Assumptions & free parameters 6 free parameters · 7 assumptions · 0 invented entities

The central claim rests on several domain assumptions rather than on new axioms. No new physical entities are introduced. Hyperparameters p, eta, alpha, K, and the number of pseudo-label sources are chosen by hand or by validation and directly shape the results.

free parameters (6)
  • alpha (KEB adjustment) = 0.3
    In Eq. (5), alpha controls how strongly the mined segmentation mask Mhat is added to pseudo-labels; Table 9 compares alpha=0.8 but no systematic sweep is reported.
  • eta (proportion factor) = 0.1
    Eq. (3) selects categories with low mean attention below eta times the average; set to 0.1 in Section 5.2 without reported sensitivity.
  • p (coverage threshold) = 98%
    Eq. (1) keeps the top categories covering 98% of instances; set in Section 5.2.
  • K-percent (dynamic augmentation selection) = 1/8
    Section 3.5 and 5.2: top K=1/8 of batch samples by KL divergence to the average attention map are selected for soft attention Mixup; no sensitivity analysis is reported.
  • number of pseudo-label sources = 2 (ML-Net + UNISAL)
    Table 8 shows the best result with two sources; using more decreases performance. This model-selection choice affects all reported results.
  • RegMixup alpha and loss weight eta = alpha=10, eta=1
    Taken from RegMixup [40] in Section 3.5; fixed hyperparameters.
assumptions (7)
  • domain assumption Pseudo-labels from natural-scene saliency models correlate with true driver attention after KEB enhancement.
    The entire unsupervised pipeline uses ML-Net, UNISAL, etc. as teachers (Section 3.1). If free-viewing saliency diverges from task-driven driver gaze in crash scenes, the targets are misleading.
  • domain assumption Mask R-CNN pretrained on MS-COCO yields reliable masks for traffic objects in driving scenes.
    KEB (Section 3.2) uses these masks without quantifying domain shift from MS-COCO to driving; errors in masks propagate into pseudo-labels.
  • domain assumption CycleGAN fog/snow and Hendrycks-style corruptions are faithful proxies for real weather and camera corruptions.
    DriverAttention-C (Section 4) is synthetic; no real-world corruption validation is provided, so robustness margins may not transfer to deployment.
  • ad hoc to paper Low mean pseudo-label attention V_ci identifies traffic-relevant categories worth boosting.
    Eq. (3) selects categories by low V_ci, then Eq. (5) raises their values. This is a self-referential selection: improvement on those categories is partly built into the targets.
  • domain assumption Pseudo-label noise follows a Gibbs/Boltzmann distribution with a per-pseudo-label scalar temperature.
    Eq. (8) is borrowed from Kendall and Gal [6]; pixel-level uncertainty maps are collapsed to a scalar mean, and no evidence supports this noise model for pseudo-label disagreement.
  • domain assumption Attention-weighted Mixup, dynamic sample selection, and random-crop RegMixup improve corruption and central-bias robustness.
    This is the core hypothesis of RoboMixup (Section 3.5), supported only empirically on BDD-A-C; the mechanism is not otherwise justified.
  • ad hoc to paper ML-Net is an appropriate reference model for mCD and Relative mCD.
    Relative mCD values in Table 3 depend on choosing ML-Net as reference; other references would change rankings, as seen when DADA has near-zero relative degradation because its clean performance is already poor.

how reviews work

0 comments
Cite this review

Pith. "Pith review of Towards Robust Unsupervised Attention Prediction in Autonomous Driving." pith.science (2026). https://pith.science/paper/57ZK2NFO

@misc{pith2026250115045,
  author       = {Pith},
  title        = {Pith review of: Towards Robust Unsupervised Attention Prediction in Autonomous Driving},
  year         = {2026},
  howpublished = {\url{https://pith.science/paper/57ZK2NFO}},
  note         = {Machine review of arXiv:2501.15045}
}
read the original abstract

Robustly predicting attention regions of interest for self-driving systems is crucial for driving safety but presents significant challenges due to the labor-intensive nature of obtaining large-scale attention labels and the domain gap between self-driving scenarios and natural scenes. These challenges are further exacerbated by complex traffic environments, including camera corruption under adverse weather, noise interferences, and central bias from long-tail distributions. To address these issues, we propose a robust unsupervised attention prediction method. An Uncertainty Mining Branch refines predictions by analyzing commonalities and differences across multiple pre-trained models on natural scenes, while a Knowledge Embedding Block bridges the domain gap by incorporating driving knowledge to adaptively enhance pseudo-labels. Additionally, we introduce RoboMixup, a novel data augmentation method that improves robustness against corruption through soft attention and dynamic augmentation, and mitigates central bias by integrating random cropping into Mixup as a regularizer. To systematically evaluate robustness in self-driving attention prediction, we introduce the DriverAttention-C benchmark, comprising over 100k frames across three subsets: BDD-A-C, DR(eye)VE-C, and DADA-2000-C. Our method achieves performance equivalent to or surpassing fully supervised state-of-the-art approaches on three public datasets and the proposed robustness benchmark, reducing relative corruption degradation by 58.8% and 52.8%, and improving central bias robustness by 12.4% and 11.4% in KLD and CC metrics, respectively. Code and data are available at https://github.com/zaplm/DriverAttention.

Figures

Figures reproduced from arXiv: 2501.15045 by the authors.

Figure 1
Figure 1. Illustration of the proposed unsupervised attention prediction in [PITH_FULL_IMAGE:figures/full_fig_p001_1.png] view at source ↗
Figure 2
Figure 2. Overview of our proposed model. We utilize pseudo-labels from models pre-trained on natural scene datasets for unsupervised training [PITH_FULL_IMAGE:figures/full_fig_p003_2.png] view at source ↗
Figure 3
Figure 3. Illustration of the knowledge embedding strategy: a) the process [PITH_FULL_IMAGE:figures/full_fig_p004_3.png] view at source ↗
Figures from the paper (5 more)
Figure 4
Figure 4. Figure 4: (a) Visual Comparison of Mixup and Our Proposed Soft Attention-based Mixup. (b) Comparison of Probability Density Distribution of the [PITH_FULL_IMAGE:figures/full_fig_p006_4.png]
Figure 5
Figure 5. Figure 5: Visualization of our generated examples, highlighting how each corruption type substantially alters the original image representation. [PITH_FULL_IMAGE:figures/full_fig_p007_5.png]
Figure 6
Figure 6. Figure 6: The visualization comparison highlights the performance of our proposed method against other methods under various corruptions. Each row [PITH_FULL_IMAGE:figures/full_fig_p011_6.png]
Figure 7
Figure 7. Figure 7: Visualization results obtained by training our proposed unsupervised model using various data augmentation methods. Each row respectively [PITH_FULL_IMAGE:figures/full_fig_p012_7.png]
Figure 8
Figure 8. Figure 8: Visualization of the distribution of normalized mean attention [PITH_FULL_IMAGE:figures/full_fig_p012_8.png]

Discussion (0). Continue with ORCID to comment.

Reference graph

Works this paper leans on

51 extracted references · 50 canonical work pages

  1. [44]

    Multi-source uncertainty mining for deep unsupervised saliency detection

    Yifan Wang, Wenbo Zhang, Lijun Wang, Ting Liu, and Huchuan Lu. Multi-source uncertainty mining for deep unsupervised saliency detection. In IEEE Conf. Comput. Vis. Pattern Recog., pages 11727–11736, 2022

    work page 2022

  2. [1]

    Medirl: Predicting the visual attention of drivers via maximum entropy deep inverse reinforcement learning

    Sonia Baee, Erfan Pakdamanian, Inki Kim, Lu Feng, Vi- cente Ordonez, and Laura Barnes. Medirl: Predicting the visual attention of drivers via maximum entropy deep inverse reinforcement learning. In Int. Conf. Comput. Vis. , pages 13178–13188, 2021

    work page 2021

  3. [2]

    ” looking at the right stuff”-guided semantic-gaze for autonomous driving

    Anwesan Pal, Sayan Mondal, and Henrik I Christensen. ” looking at the right stuff”-guided semantic-gaze for autonomous driving. In IEEE Conf. Comput. Vis. Pattern Recog., pages 11883–11892, 2020

    work page 2020

  4. [3]

    Ex- plainable object-induced action decision for autonomous vehicles

    Yiran Xu, Xiaoyin Yang, Lihang Gong, Hsuan-Chu Lin, Tz-Ying Wu, Yunsheng Li, and Nuno Vasconcelos. Ex- plainable object-induced action decision for autonomous vehicles. In IEEE Conf. Comput. Vis. Pattern Recog. , pages 9523–9532, 2020

    work page 2020

  5. [4]

    Stc-gan: Spatio-temporally coupled generative adversarial networks for predictive scene parsing

    Mengshi Qi, Yunhong Wang, Annan Li, and Jiebo Luo. Stc-gan: Spatio-temporally coupled generative adversarial networks for predictive scene parsing. IEEE T rans. Image Process., 29:5420–5430, 2020

    work page 2020

  6. [5]

    Ke- gan: Knowledge embedded generative adversarial net- works for semi-supervised scene parsing

    Mengshi Qi, Yunhong Wang, Jie Qin, and Annan Li. Ke- gan: Knowledge embedded generative adversarial net- works for semi-supervised scene parsing. In IEEE Conf. Comput. Vis. Pattern Recog. IEEE, 2019

    work page 2019

  7. [6]

    What uncertainties do we need in bayesian deep learning for computer vision? Adv

    Alex Kendall and Yarin Gal. What uncertainties do we need in bayesian deep learning for computer vision? Adv. Neural Inform. Process. Syst. , 30, 2017

    work page 2017

  8. [7]

    Predicting the driver’s focus of attention: the dr (eye) ve project

    Andrea Palazzi, Davide Abati, Francesco Solera, Rita Cuc- chiara, et al. Predicting the driver’s focus of attention: the dr (eye) ve project. IEEE T rans. Pattern Anal. Mach. Intell. , 41(7):1720–1733, 2018

    work page 2018

Show all 51 references
  1. [8]

    Predicting driver attention in critical situations

    Ye Xia, Danqing Zhang, Jinkyu Kim, Ken Nakayama, Karl Zipser, and David Whitney. Predicting driver attention in critical situations. In Asian Conf. Comput. Vis. , pages 658–

  2. [9]

    Dr (eye) ve: a dataset for attention-based tasks with applications to autonomous and assisted driving

    Stefano Alletto, Andrea Palazzi, Francesco Solera, Simone Calderara, and Rita Cucchiara. Dr (eye) ve: a dataset for attention-based tasks with applications to autonomous and assisted driving. In IEEE Conf. Comput. Vis. Pattern Recog. Worksh., pages 54–60, 2016

    2016

  3. [10]

    Dada: Driver attention prediction in driving accident scenarios

    Jianwu Fang, Dingxin Yan, Jiahuan Qiao, Jianru Xue, and Hongkai Yu. Dada: Driver attention prediction in driving accident scenarios. IEEE T rans. Intell. T ransp. Syst., 2021

    2021

  4. [11]

    Advisable learning for self-driving vehicles by internalizing observation-to-action rules

    Jinkyu Kim, Suhong Moon, Anna Rohrbach, Trevor Dar- rell, and John Canny. Advisable learning for self-driving vehicles by internalizing observation-to-action rules. In IEEE Conf. Comput. Vis. Pattern Recog. , pages 9661–9670, 2020

    2020

  5. [12]

    Deep learning-based robust positioning for all-weather autonomous driving

    Yasin Almalioglu, Mehmet Turan, Niki Trigoni, and An- drew Markham. Deep learning-based robust positioning for all-weather autonomous driving. Nature machine intel- ligence, 4(9):749–760, 2022

    2022

  6. [13]

    Benchmarking neural network robustness to common corruptions and perturbations

    Dan Hendrycks and Thomas Dietterich. Benchmarking neural network robustness to common corruptions and perturbations. Int. Conf. Learn. Represent. , 2019

    2019

  7. [14]

    When human pose estimation meets robustness: Adversarial algorithms and benchmarks

    Jiahang Wang, Sheng Jin, Wentao Liu, Weizhong Liu, Chen Qian, and Ping Luo. When human pose estimation meets robustness: Adversarial algorithms and benchmarks. In IEEE Conf. Comput. Vis. Pattern Recog. , pages 11855–11864, 2021

    2021

  8. [15]

    Contrastive adaptation network for unsu- pervised domain adaptation

    Guoliang Kang, Lu Jiang, Yi Yang, and Alexander G Hauptmann. Contrastive adaptation network for unsu- pervised domain adaptation. In IEEE Conf. Comput. Vis. Pattern Recog., pages 4893–4902, 2019

    2019

  9. [16]

    Mask r-cnn

    Kaiming He, Georgia Gkioxari, Piotr Doll ´ar, and Ross Girshick. Mask r-cnn. In Int. Conf. Comput. Vis. , pages 2961–2969, 2017

    2017

  10. [17]

    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 Eur. Conf. Comput. Vis., pages 740–755. Springer, 2014

    2014

  11. [18]

    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. In Int. Conf. Comput. Vis. , 2017

    2017

  12. [19]

    Unsupervised self-driving attention prediction via uncertainty mining and knowledge embedding

    Pengfei Zhu, Mengshi Qi, Xia Li, Weijian Li, and Huadong Ma. Unsupervised self-driving attention prediction via uncertainty mining and knowledge embedding. In Int. Conf. Comput. Vis., pages 8558–8568, 2023

    2023

  13. [20]

    Fblnet: Feed- back loop network for driver attention prediction

    Yilong Chen, Zhixiong Nan, and Tao Xiang. Fblnet: Feed- back loop network for driver attention prediction. In Int. Conf. Comput. Vis., pages 13371–13380, 2023

    2023

  14. [21]

    Tased-net: Temporally- aggregating spatial encoder-decoder network for video saliency detection

    Kyle Min and Jason J Corso. Tased-net: Temporally- aggregating spatial encoder-decoder network for video saliency detection. In Int. Conf. Comput. Vis. , pages 2394– 2403, 2019

    2019

  15. [22]

    Unified image and video saliency modeling

    Richard Droste, Jianbo Jiao, and J Alison Noble. Unified image and video saliency modeling. In Eur. Conf. Comput. Vis., pages 419–435. Springer, 2020

    2020

  16. [23]

    Deep cropping via attention box prediction and aesthetics assessment

    Wenguan Wang and Jianbing Shen. Deep cropping via attention box prediction and aesthetics assessment. In Int. Conf. Comput. Vis., pages 2186–2194, 2017

    2017

  17. [24]

    Attentive relational networks for mapping images to scene graphs

    Mengshi Qi, Weijian Li, Zhengyuan Yang, Yunhong Wang, and Jiebo Luo. Attentive relational networks for mapping images to scene graphs. In IEEE Conf. Comput. Vis. Pattern Recog. IEEE, 2019

    2019

  18. [25]

    Semantics-aware spatial-temporal binaries for cross- modal video retrieval

    Mengshi Qi, Jie Qin, Yi Yang, Yunhong Wang, and Jiebo Luo. Semantics-aware spatial-temporal binaries for cross- modal video retrieval. IEEE T rans. Image Process., 30:2989– 3004, 2021

    2021

  19. [26]

    stagnet: An attentive semantic rnn for group activity recognition

    Mengshi Qi, Jie Qin, Annan Li, Yunhong Wang, Jiebo Luo, and Luc Van Gool. stagnet: An attentive semantic rnn for group activity recognition. In Eur. Conf. Comput. Vis. springer, 2018

    2018

  20. [27]

    Bbs-net: Rgb-d salient object detection with a bifurcated backbone strategy network

    Deng-Ping Fan, Yingjie Zhai, Ali Borji, Jufeng Yang, and Ling Shao. Bbs-net: Rgb-d salient object detection with a bifurcated backbone strategy network. In Eur. Conf. Comput. Vis., pages 275–292. Springer, 2020

    2020

  21. [28]

    Pyramid grafting network for one-stage high resolution saliency detection

    Chenxi Xie, Changqun Xia, Mingcan Ma, Zhirui Zhao, Xiaowu Chen, and Jia Li. Pyramid grafting network for one-stage high resolution saliency detection. In IEEE Conf. Comput. Vis. Pattern Recog. , pages 11717–11726, 2022

    2022

  22. [29]

    Robust perception and precise segmentation for scribble-supervised rgb-d saliency detection

    Long Li, Junwei Han, Nian Liu, Salman Khan, Hisham Cholakkal, Rao Muhammad Anwer, and Fahad Shahbaz Khan. Robust perception and precise segmentation for scribble-supervised rgb-d saliency detection. IEEE T rans. Pattern Anal. Mach. Intell. , 2023

    2023

  23. [30]

    Salicon: Saliency in context

    Ming Jiang, Shengsheng Huang, Juanyong Duan, and Qi Zhao. Salicon: Saliency in context. In IEEE Conf. Comput. Vis. Pattern Recog. , pages 1072–1080, 2015

    2015

  24. [31]

    Revisiting video saliency: A large- scale benchmark and a new model

    Wenguan Wang, Jianbing Shen, Fang Guo, Ming-Ming 14 Cheng, and Ali Borji. Revisiting video saliency: A large- scale benchmark and a new model. In IEEE Conf. Comput. Vis. Pattern Recog., pages 4894–4903, 2018

    2018

  25. [32]

    A deep multi-level network for saliency prediction

    Marcella Cornia, Lorenzo Baraldi, Giuseppe Serra, and Rita Cucchiara. A deep multi-level network for saliency prediction. In Int. Conf. Pattern Recog. , pages 3488–3493. IEEE, 2016

    2016

  26. [33]

    Predicting human eye fixations via an lstm-based saliency attentive model

    Marcella Cornia, Lorenzo Baraldi, Giuseppe Serra, and Rita Cucchiara. Predicting human eye fixations via an lstm-based saliency attentive model. IEEE T rans. Image Process., 27(10):5142–5154, 2018

    2018

  27. [34]

    Mobilevit: Light-weight, general-purpose, and mobile-friendly vision transformer

    Sachin Mehta and Mohammad Rastegari. Mobilevit: Light-weight, general-purpose, and mobile-friendly vision transformer. In Int. Conf. Learn. Represent. , 2022

    2022

  28. [35]

    Laplace redux-effortless bayesian deep learning

    Erik Daxberger, Agustinus Kristiadi, Alexander Immer, Runa Eschenhagen, Matthias Bauer, and Philipp Hennig. Laplace redux-effortless bayesian deep learning. Adv. Neural Inform. Process. Syst. , 34:20089–20103, 2021

    2021

  29. [36]

    Dropout as a bayesian approximation: Representing model uncertainty in deep learning

    Yarin Gal and Zoubin Ghahramani. Dropout as a bayesian approximation: Representing model uncertainty in deep learning. In Int. Conf. Mach. Learn. , pages 1050–1059. PMLR, 2016

    2016

  30. [37]

    Multi- task learning using uncertainty to weigh losses for scene geometry and semantics

    Alex Kendall, Yarin Gal, and Roberto Cipolla. Multi- task learning using uncertainty to weigh losses for scene geometry and semantics. In IEEE Conf. Comput. Vis. Pattern Recog., pages 7482–7491, 2018

    2018

  31. [38]

    Transmix: Attend to mix for vision transformers

    Jie-Neng Chen, Shuyang Sun, Ju He, Philip HS Torr, Alan Yuille, and Song Bai. Transmix: Attend to mix for vision transformers. In IEEE Conf. Comput. Vis. Pattern Recog. , pages 12135–12144, 2022

    2022

  32. [39]

    mixup: Beyond empirical risk mini- mization

    Hongyi Zhang, Moustapha Cisse, Yann N Dauphin, and David Lopez-Paz. mixup: Beyond empirical risk mini- mization. In Int. Conf. Learn. Represent. , 2018

    2018

  33. [40]

    Using mixup as a regularizer can surprisingly improve accuracy & out-of-distribution ro- bustness

    Francesco Pinto, Harry Yang, Ser Nam Lim, Philip Torr, and Puneet Dokania. Using mixup as a regularizer can surprisingly improve accuracy & out-of-distribution ro- bustness. Adv. Neural Inform. Process. Syst., 35:14608–14622, 2022

    2022

  34. [41]

    Pyramid scene parsing network

    Hengshuang Zhao, Jianping Shi, Xiaojuan Qi, Xiaogang Wang, and Jiaya Jia. Pyramid scene parsing network. In IEEE Conf. Comput. Vis. Pattern Recog. , pages 2881–2890, 2017

    2017

  35. [42]

    A unified framework for u-net design and analysis

    Christopher Williams, Fabian Falck, George Deligiannidis, Chris C Holmes, Arnaud Doucet, and Saifuddin Syed. A unified framework for u-net design and analysis. Adv. Neural Inform. Process. Syst. , 36:27745–27782, 2023

    2023

  36. [43]

    Non-local neural networks

    Xiaolong Wang, Ross Girshick, Abhinav Gupta, and Kaim- ing He. Non-local neural networks. In IEEE Conf. Comput. Vis. Pattern Recog., pages 7794–7803, 2018

    2018

  37. [45]

    Deep residual learning for image recognition

    Kaiming He, Xiangyu Zhang, Shaoqing Ren, and Jian Sun. Deep residual learning for image recognition. In IEEE Conf. Comput. Vis. Pattern Recog. , pages 770–778, 2016

    2016

  38. [46]

    Gated2depth: Real-time dense lidar from gated images

    Tobias Gruber, Frank Julca-Aguilar, Mario Bijelic, and Felix Heide. Gated2depth: Real-time dense lidar from gated images. In Int. Conf. Comput. Vis. , 2019

    2019

  39. [47]

    Dsor: A scalable statistical filter for removing falling snow from lidar point clouds in severe winter weather

    Akhil Kurup and Jeremy Bos. Dsor: A scalable statistical filter for removing falling snow from lidar point clouds in severe winter weather. arXiv preprint arXiv:2109.07078 , 2021

    2021 arXiv

  40. [48]

    Picanet: Learning pixel-wise contextual attention for saliency de- tection

    Nian Liu, Junwei Han, and Ming-Hsuan Yang. Picanet: Learning pixel-wise contextual attention for saliency de- tection. In IEEE Conf. Comput. Vis. Pattern Recog. , pages 3089–3098, 2018

    2018

  41. [49]

    What do different evaluation metrics tell us about saliency models? IEEE T rans

    Zoya Bylinskii, Tilke Judd, Aude Oliva, Antonio Torralba, and Fr ´edo Durand. What do different evaluation metrics tell us about saliency models? IEEE T rans. Pattern Anal. Mach. Intell., 41(3):740–757, 2018

    2018

  42. [50]

    Pytorch: An imperative style, high-performance deep learning library

    Adam Paszke, Sam Gross, Francisco Massa, Adam Lerer, James Bradbury, Gregory Chanan, Trevor Killeen, Zeming Lin, Natalia Gimelshein, Luca Antiga, et al. Pytorch: An imperative style, high-performance deep learning library. Adv. Neural Inform. Process. Syst. , 32, 2019

    2019

  43. [51]

    Distill- ing the knowledge in a neural network

    Geoffrey Hinton, Oriol Vinyals, and Jeffrey Dean. Distill- ing the knowledge in a neural network. In Adv. Neural Inform. Process. Syst. Worksh. , 2015

    2015

Pith tools

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