pith. sign in

arxiv: 1906.09806 · v1 · pith:WA2QZIRSnew · submitted 2019-06-24 · 💻 cs.CV

Saliency Detection With Fully Convolutional Neural Network

Hooman Misaghi , Reza Askari Moghadam , Ali Mahmoudi , Kurosh Madani This is my paper

Pith reviewed 2026-05-25 17:50 UTC · model grok-4.3

classification 💻 cs.CV
keywords saliency detectionfully convolutional networkVGG-16image processingpretrained weightsconvolutional neural network
0
0 comments X

The pith

A fully convolutional network built from VGG-16 layers marks salient regions in images.

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

The authors propose adapting part of a pretrained VGG-16 network into a fully convolutional architecture to produce saliency maps. The approach rests on the established practice of transferring weights from classification models to new image tasks. If the transfer succeeds, saliency detection becomes possible without training an entire network from random initialization and without designing a task-specific backbone from scratch.

Core claim

The paper claims that a fully convolutional neural network using a portion of VGG-16 can perform saliency detection by reusing the pretrained weights of the classification network.

What carries the argument

The fully convolutional neural network that takes a subset of VGG-16 layers and repurposes them to output a saliency map instead of class scores.

If this is right

  • Saliency maps can be generated by adapting existing classification networks rather than building new ones.
  • Pretrained weights from image classification improve performance on the saliency task.
  • The same reuse strategy could apply to other early-stage image processing steps that follow saliency detection.
  • Training time and data requirements for saliency models decrease when starting from VGG-16 weights.

Where Pith is reading between the lines

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

  • The same layer-reuse pattern might extend to related tasks such as edge detection or object proposal generation.
  • Performance would likely vary with the choice of which VGG-16 layers are retained versus replaced.
  • Real-time applications could benefit if the resulting network runs at interactive speeds on standard hardware.

Load-bearing premise

Reusing layers and weights from a classification network will produce accurate saliency maps without further architectural changes or task-specific validation.

What would settle it

Evaluating the network outputs against human-annotated saliency ground truth on standard image datasets and measuring overlap metrics such as precision-recall or F-measure.

Figures

Figures reproduced from arXiv: 1906.09806 by Ali Mahmoudi, Hooman Misaghi, Kurosh Madani, Reza Askari Moghadam.

Figure 1
Figure 1. Figure 1: Proposed Network’s architecture activations the results will never be exact 0. Training was done for 20 epochs with batch size of 20 on HKU-IS dataset. VGG layers were freezed and no update operation was applied on them. IV. RESULTS AND EVALUATION In this network an FCNN(Fully Convolutional Neural Network) architecture based on VGG 16 has been proposed. The results of the trained network are presented in t… view at source ↗
Figure 3
Figure 3. Figure 3: (a) Original image (b) Saliency map from the [PITH_FULL_IMAGE:figures/full_fig_p003_3.png] view at source ↗
Figure 2
Figure 2. Figure 2: Precision-Recall curve [PITH_FULL_IMAGE:figures/full_fig_p003_2.png] view at source ↗
read the original abstract

Saliency detection is an important task in image processing as it can solve many problems and it usually is the first step in for other processes. Convolutional neural networks have been proved to be very effective on several image processing tasks such as classification, segmentation, semantic colorization and object manipulation. Besides, using the weights of a pretrained networks is a common practice for enhancing the accuracy of a network. In this paper a fully convolutional neural network which uses a part of VGG-16 is proposed for saliency detection in images.

Editorial analysis

A structured set of objections, weighed in public.

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

Referee Report

1 major / 1 minor

Summary. The manuscript proposes a fully convolutional neural network for saliency detection that reuses a portion of the pretrained VGG-16 classification network, motivated by the effectiveness of CNNs and transfer learning for image processing tasks.

Significance. Adapting pretrained classification backbones to saliency detection is a standard technique, but the manuscript supplies neither architectural specifics beyond the VGG-16 reference, nor any training protocol, datasets, baselines, or quantitative results. Consequently, even if the high-level idea holds, the work adds no verifiable contribution or falsifiable prediction.

major comments (1)
  1. [Abstract] Abstract: the central claim that the proposed network is effective for saliency detection cannot be evaluated because the text contains no architecture diagram, layer specifications, loss function, training details, evaluation metrics, or experimental results on any dataset.
minor comments (1)
  1. [Abstract] Abstract, line 3: 'first step in for other processes' contains a grammatical error.

Simulated Author's Rebuttal

1 responses · 0 unresolved

We thank the referee for the review. We agree that the submitted manuscript is a brief proposal lacking the details needed for evaluation and will revise it substantially.

read point-by-point responses

  1. Referee: [Abstract] Abstract: the central claim that the proposed network is effective for saliency detection cannot be evaluated because the text contains no architecture diagram, layer specifications, loss function, training details, evaluation metrics, or experimental results on any dataset.

    Authors: We agree the manuscript provides only a high-level proposal and contains none of the listed elements. The text proposes the network but does not claim or demonstrate effectiveness with evidence. In revision we will add an architecture diagram, layer-by-layer specifications reusing VGG-16, the loss function, training protocol, chosen datasets, evaluation metrics, and quantitative results against baselines. revision: yes

Circularity Check

0 steps flagged

No significant circularity; proposal is architectural description only

full rationale

The paper contains no equations, derivations, or load-bearing claims that reduce to self-definition, fitted inputs, or self-citation chains. The central statement is simply that an FCN reusing part of VGG-16 is proposed for saliency detection, with no internal logical steps or predictions that could be circular by construction. The provided abstract and description confirm this is a straightforward network proposal without any mathematical reduction or uniqueness theorem invoked.

Axiom & Free-Parameter Ledger

0 free parameters · 0 axioms · 0 invented entities

No mathematical content, free parameters, or new entities are introduced in the abstract; the work rests on the standard assumption that pretrained classification weights transfer usefully to pixel-wise prediction.

pith-pipeline@v0.9.0 · 5612 in / 885 out tokens · 21002 ms · 2026-05-25T17:50:48.298973+00:00 · methodology

discussion (0)

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

Reference graph

Works this paper leans on

14 extracted references · 14 canonical work pages · 2 internal anchors

  1. [1]

    Salient region detection and segmentation,

    R. Achanta, F. Estrada, P. Wils, and S. S¨usstrunk , “Salient region detection and segmentation,” in International conference on computer vision systems , pp. 66–75, Springer, 2008

    work page 2008

  2. [2]

    Efficient graph - based image segmentation,

    P. F. Felzenszwalb and D. P. Huttenlocher, “Efficient graph - based image segmentation,” International journal of computer vision, vol. 59, no. 2, pp. 167–181, 2004

    work page 2004

  3. [3]

    Mean shift: A robust approach toward feature space analysis,

    D. Comaniciu and P. Meer, “Mean shift: A robust approach toward feature space analysis,” IEEE Transactions on pattern analysis and machine intelligence , vol. 24, no. 5, pp. 603 – 619, 2002

    work page 2002

  4. [4]

    A machine learning based intelligent vision system for autonomous object detection and recognition,

    D. M. Ram´ık, C. Sabourin, R. Moreno, and K. Madani, “A machine learning based intelligent vision system for autonomous object detection and recognition,” Applied intelligence, vol. 40, no. 2, pp. 358–375, 2014

    work page 2014

  5. [5]

    Deep learning for visual understanding: A review,

    Y. Guo, Y. Liu, A. Oerlemans, S. Lao, S. Wu, and M. S. Lew, “Deep learning for visual understanding: A review,” eurocomputing, vol. 187, pp. 27–48, 2016

    work page 2016

  6. [7]

    Visual saliency based on multiscale deep features,

    G. Li and Y. Yu, “Visual saliency based on multiscale deep features,” in Proceedings of the IEEE conference on computer vision and pattern recognition , pp. 5455 –5463, 2015

    work page 2015

  7. [8]

    Convolutional neural network for saliency detection in images,

    H. Misaghi, R. A. Moghadam, and K. Madani, “Convolutional neural network for saliency detection in images,” in Fuzzy and Intelligent Systems (CFIS), 2018 6th Iranian Joint Congress on, pp. 17–19, IEEE, 2018

    work page 2018

  8. [9]

    Saliency detection by multi -context deep learning,

    R. Zhao, W. Ouyang, H. Li, and X. Wang, “Saliency detection by multi -context deep learning,” in Proceedings of the IEEE Conference on Computer Vision and Pattern Recognition, pp. 1265–1274, 2015

    work page 2015

  9. [10]

    Very Deep Convolutional Networks for Large-Scale Image Recognition

    K. Simonyan and A. Zisserman, “Very deep convolutional networks for large -scale image recognition,” arXiv preprint arXiv:1409.1556, 2014

    work page internal anchor Pith review Pith/arXiv arXiv 2014

  10. [11]

    Global contrast based salient region detection,

    M.-M. Cheng, N. J. Mitra, X. Huang, P. H. Torr, and S .-M. Hu, “Global contrast based salient region detection,” IEEE Transactions on Pattern Analysis and Machine Intelligence , vol. 37, no. 3, pp. 569–582, 2015

    work page 2015

  11. [12]

    Deep contrast learning for salient object detection,

    G. Li and Y. Yu, “Deep contrast learning for salient object detection,” in Proceedings of the IEEE Conference on Computer Vision and Pattern Recognition , pp. 478 –487, 2016

    work page 2016

  12. [13]

    Adam: A Method for Stochastic Optimization

    D. P. Kingma and J. Ba, “Adam: A method for stochastic optimization,” arXiv preprint arXiv:1412.6980, 2014

    work page internal anchor Pith review Pith/arXiv arXiv 2014

  13. [14]

    Deeply supervi sed salient object detection with short connections,

    Q. Hou, M. -M. Cheng, X. Hu, A. Borji, Z. Tu, and P. Torr, “Deeply supervi sed salient object detection with short connections,” IEEE TPAMI, 2018

    work page 2018

  14. [15]

    Pisa: Pixelwise image saliency by aggregating complementary appearance contrast measures with edge -preserving coherence,

    K. Wang, L. Lin, J. Lu, C. Li, and K. Shi, “Pisa: Pixelwise image saliency by aggregating complementary appearance contrast measures with edge -preserving coherence,” IEEE Transactions on Image Processing, vol. 24, no. 10, pp. 3019 – 3033, 2015. 

    work page 2015