Pith. sign in

REVIEW 4 major objections 9 minor 22 references

Advancing TDFN: Precise Fixation Point Generation Using Reconstruction Differences

T0 review · 4 major / 9 minor · reviewed 2026-08-10 · deepseek-v4-flash

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

Pith's one-line read TDFN can train its fixation-point generator against the reconstruction error map |ReconImage − InputImage|, and doing so yields pixel-accurate fixations that improve classification accuracy and cut the number of fixations needed.

desk verdict Simple self-supervised fixation target from reconstruction error beats their RL baseline on MNIST, but the evidence is too thin to support the general claim. read the letter →

arxiv 2501.15603 v1 pith:UNDZ2Y4I submitted 2025-01-26 cs.CV

Shuguang Wang , Yuanjing Wang This is my paper · ORCID
classification cs.CV MSC 68T0768T45
keywords fixationpointgenerationTask-DrivenNetworkreconstructiondifferencesaliencymapdynamicpatchselectionTransformerimageclassificationMNIST
verification ladder T0 review T1 audit T2 compute T3 formal

The pith

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

The reading

The paper replaces the reinforcement-learning-based fixation point selector inside the Task-Driven Fixation Network (TDFN), a Transformer that classifies an image from a low-resolution global view plus a few high-resolution local crops centered on 'fixation points,' with a supervised regression. The new generator is trained to output a saliency map that matches the absolute pixel-wise difference between the network's internally reconstructed image and the actual input, and the next fixation is placed where that difference is largest. On MNIST this yields fixation points at stroke endpoints, corners, and intersections, and it improves classification accuracy at every fixation budget tested—two fixations rise from 84.20% to 94.76%, eight from 96.11% to 99.19%. The method also lets TDFN stop early: with an average of 1.57 fixations it reaches 99.29% accuracy while covering under 10% of the image. If it holds beyond MNIST, this gives a cheap, differentiable way to do dynamic, content-adaptive patching in vision Transformers.

What carries the argument

The load-bearing object is the reconstruction error map |ReconImage − InputImage| (Eq. 7). The reconstructor predicts a full-resolution image from the available low-resolution plus ROI inputs; the fixation generator is trained to predict this error map; and the next fixation is the saliency map's argmax. The paper motivates this with a Taylor expansion identity (Eq. 1) that ties quick loss reduction to large differences between the input and a template, with the reconstructed image standing in for the template. The machinery consists of three jointly trained modules—classifier, reconstructor, and fixation generator—with total loss ClassLoss + α·ReconLoss + β·FixationLoss, using α = 0.2 and β = 0.1.

What would settle it

A straightforward test: on a variant of MNIST where a high-contrast irrelevant mark is added to the background, the reconstruction error map will concentrate on the mark. If FPG2 fixates there and accuracy drops relative to FPG1, the reconstruction-error proxy is not tracking classification-relevant content.

Watch

Extended reading notes

Core claim

Starting from a Taylor expansion of the loss around a 'template' input x*, the paper argues that the fastest way to reduce loss is to attend where the derivative L'(x) is large and where (x* − x) is large. Since TDFN cannot compute the derivative at inference time, the paper substitutes a network-internal reconstruction of the input for x*, turning (x* − x) into the computable reconstruction error map |ReconImage − InputImage|. The fixation generator is trained with mean squared error against this error map (Eq. 6), so the argmax of its saliency map is the next fixation point. The paper reports that on 32×32 MNIST this 'FPG2' generator produces fixations at informative locations and beats the prior reinforcement-learning 'FPG1' generator in both fixed-budget and early-termination comparisons.

Load-bearing premise

The paper assumes that the pixel-wise difference between the reconstructed image and the input marks the places where an extra high-resolution look will most help classification, even though the reconstruction is trained to match pixel intensities rather than to improve classification.

Editorial extensions

If this is right

  • Pixel-level fixation generation becomes a supervised regression problem instead of a reinforcement-learning problem, eliminating the need to scale RL to the full image grid.
  • On MNIST, FPG2 improves classification accuracy at every fixation count tested, e.g., 84.20% to 94.76% with two fixations and 96.11% to 99.19% with eight.
  • With dynamic termination, FPG2 averages 1.57 fixations to reach 99.29% accuracy at an MCP threshold of 0.995, covering only 9.84% of the image.
  • The improved fixations let TDFN act as a dynamic patching mechanism: fine-grained attention where differences are large, coarse elsewhere, at lower computational cost than fixed fine-grained ViT patches.

Reading between the lines

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

  • Extension beyond the paper: the reconstruction-error proxy may not align with classification-relevant content. A direct test would compare fixation targets from reconstruction error against class-saliency maps from occluding pixels; if the overlap is no better than chance, the accuracy gain on MNIST may come from digits exactly occupying high-error regions.
  • Extension: on natural images, background clutter will produce large reconstruction errors irrelevant to the class, so the method may require a reconstruction trained with a classification-aware term or a saliency gating mechanism.
  • Extension: the same target could be used for other tasks TDFN might support, such as regression or detection, since the reconstruction error map does not depend on the task head.
  • Testable: increasing the reconstruction-loss weight α at the expense of classification loss should produce sharper saliency maps but could degrade accuracy if reconstruction and classification goals are not aligned.
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 / 9 minor

Summary. This paper proposes FPG2, a supervised replacement for the reinforcement-learning-based fixation point generator (FPG1) in the Task-Driven Fixation Network (TDFN). The core idea is to train a fixation generator to predict the absolute difference between an internally reconstructed image and the input image (Eqs. 6–7), and to select the argmax of the resulting saliency map as the next fixation point. The method is motivated by a Taylor-expansion argument in §3.1 in which the reconstructed image is identified with a loss-minimizing 'template' x*. Experiments on MNIST (§4) report that FPG2 achieves higher classification accuracy than FPG1 for the same number of fixations and higher accuracy with fewer fixation steps under dynamic termination.

Significance. If its central premise is accepted — that the reconstruction error map is a task-relevant saliency signal — the paper's engineering contribution is genuine: it replaces an unstable RL procedure with a straightforward supervised MSE regression (Eq. 6), enables fixation selection at full 32×32 resolution, and reuses TDFN's existing reconstruction module. The paper is commendably honest, stating in the Conclusion that the method 'is not the optimal solution' and is tailored to TDFN. The dynamic-termination analysis in Table 2 is a useful addition, and the premise is falsifiable via ablations. However, the theoretical justification is at present only a heuristic, the load-bearing identification of reconstruction error with classification-relevant difference is unproven, and the empirical support is limited to single-run MNIST results against the authors' own baseline. The paper releases no code; its strengths are the clarity and simplicity of the proposal rather than the strength of the evidence.

major comments (4)
  1. [§3.1, Eq. (1)] The Taylor-expansion argument does not, as written, support the fixation-selection criterion. TDFN's classifier loss is a function of the low-resolution global input and the sequence of selected ROIs, not of the full-resolution image x; the ROI selection is discrete (argmax in Eq. 5), so the gradient L'(x) with respect to the full image is not the derivative of the actually optimized function. The step 'Assuming L(x*) is very small and can be ignored' also conflates the minimum loss value with the truncation error of the expansion. As it stands, Eq. (1) yields at most a heuristic; the authors should either label it as such and support it empirically, or derive the criterion from the classifier's true inputs.
  2. [§3.2–§3.3, Eqs. (7), (10)] The substitution of the reconstructed image for the template x* is unproven and load-bearing. The reconstructor is trained only with pixel-wise MSE (Eq. 10), which measures pixel fidelity; a reconstructor can have small MSE while being wrong at classification-critical pixels, and its largest errors may lie on textures, edges, or background clutter with no class evidence. Since the training target ReconErrMap (Eq. 7) is produced by the network's own reconstruction module, FPG2's quality is entirely inherited from the unproven task-relevance of that internally generated map. The paper should test the premise directly by replacing ReconErrMap with alternative targets (gradient-based saliency, edge magnitude, random, or center fixations) under identical training and showing that the reconstruction-difference target is what drives the gains. This ablation is essential on MNIST, where digit strokes dominate both reconstruction error and class-relevant pixels.
  3. [§4.3–§4.4, Tables 1–2] The experimental evidence is too thin for the 'significantly enhances' claims. All accuracies come from a single run, with no error bars, no variance over seeds, and no significance test. The only comparator is FPG1, the authors' own RL method; there are no trivial baselines such as random fixations, center fixation, or a classical saliency method. Since FPG1 selects on a 4×4 grid while FPG2 selects on the full 32×32 image, the gains could partly reflect the finer decision grid rather than the proposed target. In Table 2, matched MCP thresholds do not equate operating points because FPG1 and FPG2 operate at very different coverage levels (e.g., 38.54% versus 7.78% at MCP 0.99), and the abstract's phrase 'achieve a predefined accuracy level' does not match the protocol, which terminates on the classifier's maximum probability rather than on achieved accuracy. Multi-seed results, at least one non-MNIST dataset, and fixed operating points are needed.
  4. [§4.2, Fig. 3] The claim of 'highly accurate fixation points' is operationalized only through downstream classification accuracy and subjective visualization; no quantitative measure of fixation precision is provided. Without ground-truth fixation annotations or digit-part masks, statements such as fixations being 'more accurately located at stroke endings, corners, intersections, and edges' (§4.2) cannot be checked. The authors should either quantify precision (e.g., distance to digit-stroke pixels, or agreement with human fixations) or rephrase the claim as task-performance improvement rather than spatial precision.
minor comments (9)
  1. [Title/page 1] The running title on the first page reads 'P RECISE FIXATION POINT GENERATION'; it should read 'PRECISE FIXATION POINT GENERATION'.
  2. [§3.1, Eq. (1)] The expansion L(x*) ≈ L(x) + L'(x)^T (x* − x) should define the gradient with respect to the image and state that the second-order term is omitted, so the approximation is transparent.
  3. [§4.1] In §4.1, 'All images were normalized to dimensions of 32 × 32 pixels' should specify the resizing or padding operation and any value normalization (e.g., mean and standard deviation), since these directly affect both the reconstruction-error target and the classification accuracy.
  4. [§4.1] Section 4.1 reports only the chosen values α = 0.2 and β = 0.1 for Eq. (8); the optimizer, learning rate, batch size, number of epochs, and any sensitivity analysis for α and β are missing.
  5. [§3.3, Eq. (6)] In Eq. (6), it is unclear whether the SaliencyMap is normalized (e.g., by a sigmoid) before the MSE with ReconErrMap is computed; a sentence on the scaling of the two maps would rule out trivial scale artifacts in FixationLoss.
  6. [§4.4, Table 2] In Table 2, the rows for MCP thresholds 0.993 and 0.995 show '–' for FPG1; the paper should state explicitly why FPG1 does not reach these thresholds (e.g., saturation of its maximum classification probability).
  7. [References] The in-text citation 'Yulin Xie et al., 2013' should be 'Xie et al., 2013', and the DeepGaze II reference entry contains a typo ('arXiv.1610.01563' should be 'arXiv:1610.01563').
  8. [§2] The statement in §2 that the fixed patch schema of ViT is 'impractical' for pixel-level segmentation is stronger than needed for the paper's argument and could be softened.
  9. [§5/Abstract] The Conclusion states that the method 'is not the optimal solution' and is tailored to TDFN; this honest limitation should be reflected in the abstract, which currently claims a general ability to achieve 'highly accurate fixation points' without qualification.

Circularity Check

2 steps flagged · score 6.0 of 10

Fixation 'prediction' reduces to fitting the network's own reconstruction-error map; the Taylor template substitution is definitional rather than derived.

  1. self definitional [Section 3.1-3.3, Eqs. (1), (3), (6)-(7)]
    "By substituting the reconstructed image for x∗, the computation of (x∗ − x) becomes significantly simpler. ... In our method, this reconstructed image serves as the template x∗ in the earlier formulation. ... F ixationLoss= M SE(SaliencyM ap, ReconErrM ap) (6) ... ReconErrM ap= |ReconImage − InputImage | (7)"

    Equation (1) defines x∗ as an optimal input that minimizes the classification loss, so (x∗ − x) is a classification-relevant perturbation. The paper then simply lets the network's own reconstruction output serve as x∗. But ReconImage is trained only by pixel-wise MSE to the input (Eq. 10), not by any condition involving classification loss. Thus |ReconImage − InputImage| is not the (x∗ − x) of Eq. (1); it is an internally generated error map. The claim that selecting its largest entries reduces classification loss follows only because the map was renamed x∗ − x by definition, making the 'first-principles' justification self-definitional.

  2. fitted input called prediction [Section 3.3, Eqs. (4)-(7); Section 4.2]
    "Instead of the reinforcement learning approach used in the original work, this paper trains the fixation point generator to predict the difference between the reconstructed image and the actual input image. ... F ixationLoss= M SE(SaliencyM ap, ReconErrM ap) (6) ... ReconErrM ap= |ReconImage − InputImage | (7) ... the Fixation Generator produces salient points at stroke endpoints, corners, intersections, and edges, aligning well with our expectations."

    The generator is trained with MSE to reproduce the reconstruction-error map that the same network computes from its own reconstruction output. If training converges, SaliencyMap ≈ ReconErrMap, so the 'predicted' fixation point is simply the argmax of an internally computed map. There is no independent fixation ground truth against which 'highly accurate fixation points' is tested; the reported localization quality is an evaluation of how well the generator copies an internally defined target. The fixation 'prediction' is therefore forced by the training objective rather than being an independent result, although the downstream classification-accuracy numbers remain externally measured.

full rationale

The paper's downstream classification-accuracy comparisons (Tables 1 and 2) are external and are not themselves circular: FPG2 is compared with FPG1 on MNIST labels, and the reported accuracy and average fixation steps are valid empirical measurements. However, the core novelty of the paper, the fixation point generation method, is justified in a circular way. The Taylor-expansion argument in Section 3.1 defines x∗ as an optimal input for the classification loss, but Section 3.2 substitutes the network's own reconstruction output for x∗, which is trained only by pixel-wise MSE. The resulting target ReconErrMap is therefore an internally generated quantity, not the (x∗ − x) of Eq. (1). Then Section 3.3 trains the fixation generator to fit this same internal map and selects its argmax, so the 'prediction' of fixation points reduces by construction to copying the reconstruction-error map. The authors partly concede this in their limitation note: 'the proposed method is not the optimal solution. Instead, it is tailored to meet the specific requirements of TDFN.' No external saliency or human-fixation benchmark is used, and no error bars or baselines such as center fixation or random fixation are reported, so the accuracy of the fixation localization itself is not independently validated. Because the central localization claim reduces to fitting an internally generated target, while the external classification gains remain meaningful, the circularity score is 6 rather than higher.

Assumptions & free parameters 2 free parameters · 2 assumptions · 0 invented entities

The ledger shows that the method rests on domain assumptions about the relationship between reconstruction and classification, and on two hand-chosen loss weights. No new physical entities are introduced.

free parameters (2)
  • alpha = 0.2
    Weight for reconstruction loss in Eq. (8). Chosen by hand, no sensitivity reported.
  • beta = 0.1
    Weight for fixation loss in Eq. (8). Chosen by hand, no sensitivity reported.
assumptions (2)
  • domain assumption The reconstruction error map is a sufficient proxy for the classification-relevant difference (x* - x) in the Taylor expansion (Eq. 1).
    The method substitutes the reconstructed image for the optimal template x* and uses the absolute difference as the fixation target. This identification is asserted, not derived, in Section 3.1 and Section 3.3.
  • domain assumption Selecting the pixel with maximal predicted reconstruction error is the best way to choose the next fixation point.
    The greedy argmax rule (Eq. 5) is used without comparing to other selection policies or multi-point planning.

how reviews work

0 comments
Cite this review

Pith. "Pith review of Advancing TDFN: Precise Fixation Point Generation Using Reconstruction Differences." pith.science (2026). https://pith.science/paper/UNDZ2Y4I

@misc{pith2026250115603,
  author       = {Pith},
  title        = {Pith review of: Advancing TDFN: Precise Fixation Point Generation Using Reconstruction Differences},
  year         = {2026},
  howpublished = {\url{https://pith.science/paper/UNDZ2Y4I}},
  note         = {Machine review of arXiv:2501.15603}
}
read the original abstract

Wang and Wang (2025) proposed the Task-Driven Fixation Network (TDFN) based on the fixation mechanism, which leverages low-resolution information along with high-resolution details near fixation points to accomplish specific visual tasks. The model employs reinforcement learning to generate fixation points. However, training reinforcement learning models is challenging, particularly when aiming to generate pixel-level accurate fixation points on high-resolution images. This paper introduces an improved fixation point generation method by leveraging the difference between the reconstructed image and the input image to train the fixation point generator. This approach directs fixation points to areas with significant differences between the reconstructed and input images. Experimental results demonstrate that this method achieves highly accurate fixation points, significantly enhances the network's classification accuracy, and reduces the average number of required fixations to achieve a predefined accuracy level.

Figures

Figures reproduced from arXiv: 2501.15603 by the authors.

Figure 1
Figure 1. TDFN architecture. 3 Method 3.1 Fixation Point Selection from the Perspective of the Loss Function From the perspective of the loss function, the selection of fixation points can be analyzed mathematically. Assuming the network parameters are fixed, the loss function L can be expressed as a function of the input image x, i.e., L(x). Let us further assume that there exists an optimal input x ∗ , called the template, … view at source ↗
Figure 2
Figure 2. Different maps output by TDFN. The first column shows the original input image, the second column [PITH_FULL_IMAGE:figures/full_fig_p005_2.png] view at source ↗
Figure 3
Figure 3. Visualization of Fixation Points by FPG2. The first column shows the original input images. The second [PITH_FULL_IMAGE:figures/full_fig_p006_3.png] view at source ↗

Discussion (0). Continue with ORCID to comment.

Reference graph

Works this paper leans on

22 extracted references · 14 canonical work pages

  1. [1]

    Quantifying Attention Flow in Transformers , May 2020

    Samira Abnar and Willem Zuidema. Quantifying Attention Flow in Transformers , May 2020. arXiv:2005.00928

    arXiv 2020

  2. [2]

    An Image is Worth 16x16 Words : Transformers for Image Recognition at Scale , June 2021

    Alexey Dosovitskiy, Lucas Beyer, Alexander Kolesnikov, Dirk Weissenborn, Xiaohua Zhai, Thomas Unterthiner, Mostafa Dehghani, Matthias Minderer, Georg Heigold, Sylvain Gelly, Jakob Uszkoreit, and Neil Houlsby. An Image is Worth 16x16 Words : Transformers for Image Recognition at Scale , June 2021. arXiv:2010.11929

    arXiv 2021

  3. [3]

    Decision- Theoretic Saliency : Computational Principles , Biological Plausibility , and Implications for Neurophysiology and Psychophysics

    Dashan Gao and Nuno Vasconcelos. Decision- Theoretic Saliency : Computational Principles , Biological Plausibility , and Implications for Neurophysiology and Psychophysics . Neural Computation, 21 0 (1): 0 239--271, January 2009

    work page 2009

  4. [4]

    SALICON : Reducing the Semantic Gap in Saliency Prediction by Adapting Deep Neural Networks

    Xun Huang, Chengyao Shen, Xavier Boix, and Qi Zhao. SALICON : Reducing the Semantic Gap in Saliency Prediction by Adapting Deep Neural Networks . In 2015 IEEE International Conference on Computer Vision ( ICCV ) , pages 262--270, Santiago, Chile, December 2015. IEEE. ISBN 978-1-4673-8391-2

    work page 2015

  5. [5]

    L. Itti, C. Koch, and E. Niebur. A model of saliency-based visual attention for rapid scene analysis. IEEE Transactions on Pattern Analysis and Machine Intelligence, 20 0 (11): 0 1254--1259, November 1998

    work page 1998

  6. [6]

    Koch and S

    C. Koch and S. Ullman. Shifts in selective visual attention: Towards the underlying neural circuitry. Human Neurobiology, 4 0 (4): 0 219--227, 1985

    work page 1985

  7. [7]

    Deep Gaze I : Boosting Saliency Prediction with Feature Maps Trained on ImageNet , April 2015

    Matthias K \"u mmerer, Lucas Theis, and Matthias Bethge. Deep Gaze I : Boosting Saliency Prediction with Feature Maps Trained on ImageNet , April 2015. arXiv:1411.1045

    arXiv 2015

  8. [8]

    Matthias K \"u mmerer, Thomas S. A. Wallis, and Matthias Bethge. DeepGaze II : Reading fixations from deep features trained on object recognition, October 2016. arXiv.1610.01563

    arXiv 2016

Show all 22 references
  1. [9]

    Oliva, A

    A. Oliva, A. Torralba, M.S. Castelhano, and J.M. Henderson. Top-down control of visual attention in object detection. In Proceedings 2003 International Conference on Image Processing ( Cat . No . 03CH37429 ) , volume 1, pages I--253--6, Barcelona, Spain, 2003. IEEE. ISBN 978-0...

    2003

  2. [10]

    O'Connor, Jordi Torres, Elisa Sayrol, and Xavier Giro-i-Nieto

    Junting Pan, Cristian Canton Ferrer, Kevin McGuinness, Noel E. O'Connor, Jordi Torres, Elisa Sayrol, and Xavier Giro-i-Nieto . SalGAN : Visual Saliency Prediction with Generative Adversarial Networks , July 2018. arXiv:1701.01081

    2018 arXiv

  3. [11]

    Peters and Laurent Itti

    Robert J. Peters and Laurent Itti. Beyond bottom-up: Incorporating task-dependent influences into a computational model of spatial attention. In 2007 IEEE Conference on Computer Vision and Pattern Recognition , pages 1--8, Minneapolis, MN, USA, June 2007. IEEE. ISBN 978-1-4244-1179-5

    2007

  4. [12]

    Deep Inside Convolutional Networks : Visualising Image Classification Models and Saliency Maps , April 2014

    Karen Simonyan, Andrea Vedaldi, and Andrew Zisserman. Deep Inside Convolutional Networks : Visualising Image Classification Models and Saliency Maps , April 2014. arXiv:1312.6034

    2014 arXiv

  5. [13]

    Treisman and Garry Gelade

    Anne M. Treisman and Garry Gelade. A feature-integration theory of attention. Cognitive Psychology, 12 0 (1): 0 97--136, January 1980

    1980

  6. [14]

    Gomez, Lukasz Kaiser, and Illia Polosukhin

    Ashish Vaswani, Noam Shazeer, Niki Parmar, Jakob Uszkoreit, Llion Jones, Aidan N. Gomez, Lukasz Kaiser, and Illia Polosukhin. Attention Is All You Need , June 2017. arXiv:1706.03762

    2017 arXiv

  7. [15]

    Large- Scale Optimization of Hierarchical Features for Saliency Prediction in Natural Images

    Eleonora Vig, Michael Dorr, and David Cox. Large- Scale Optimization of Hierarchical Features for Saliency Prediction in Natural Images . In 2014 IEEE Conference on Computer Vision and Pattern Recognition , pages 2798--2805, Columbus, OH, USA, June 2014. IEEE. ISBN 978-1-4799-5118-5

    2014

  8. [16]

    Task- Driven Fixation Network : An Efficient Architecture with Fixation Selection , January 2025

    Shuguang Wang and Yuanjing Wang. Task- Driven Fixation Network : An Efficient Architecture with Fixation Selection , January 2025. arXiv:2501.01548

    2025 arXiv

  9. [17]

    Inferring Salient Objects from Human Fixations

    Wenguan Wang, Jianbing Shen, Xingping Dong, Ali Borji, and Ruigang Yang. Inferring Salient Objects from Human Fixations . IEEE Transactions on Pattern Analysis and Machine Intelligence, 42 0 (8): 0 1913--1927, August 2020

    1913

  10. [18]

    Pyramid Vision Transformer : A Versatile Backbone for Dense Prediction without Convolutions , August 2021

    Wenhai Wang, Enze Xie, Xiang Li, Deng-Ping Fan, Kaitao Song, Ding Liang, Tong Lu, Ping Luo, and Ling Shao. Pyramid Vision Transformer : A Versatile Backbone for Dense Prediction without Convolutions , August 2021. arXiv:2102.12122

    2021 arXiv

  11. [19]

    Review of Visual Saliency Prediction : Development Process from Neurobiological Basis to Deep Models

    Fei Yan, Cheng Chen, Peng Xiao, Siyu Qi, Zhiliang Wang, and Ruoxiu Xiao. Review of Visual Saliency Prediction : Development Process from Neurobiological Basis to Deep Models . Applied Sciences, 12 0 (1): 0 309, December 2021

    2021

  12. [20]

    Bayesian Saliency via Low and Mid Level Cues

    Yulin Xie , Huchuan Lu , and Ming-Hsuan Yang . Bayesian Saliency via Low and Mid Level Cues . IEEE Transactions on Image Processing, 22 0 (5): 0 1689--1698, May 2013

    2013

  13. [21]

    Tong, Tim K

    Lingyun Zhang, Matthew H. Tong, Tim K. Marks, Honghao Shan, and Garrison W. Cottrell. SUN : A Bayesian framework for saliency using natural statistics. Journal of Vision, 8 0 (7): 0 32, December 2008

    2008

  14. [22]

    Learning Deep Features for Discriminative Localization

    Bolei Zhou, Aditya Khosla, Agata Lapedriza, Aude Oliva, and Antonio Torralba. Learning Deep Features for Discriminative Localization . In 2016 IEEE Conference on Computer Vision and Pattern Recognition ( CVPR ) , pages 2921--2929, Las Vegas, NV, USA, June 2016. IEEE. ISBN 978-...

    2016

Pith tools

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