Logo
ScienceStack
Table of Contents
Fetching ...
Sign In

BRI3L: A Brightness Illusion Image Dataset for Identification and Localization of Regions of Illusory Perception

Aniket Roy, Anirban Roy, Soma Mitra, Kuntal Ghosh

TL;DR

BRI3L introduces a large-scale, labeled dataset of 22,366 images capturing five brightness Illusions with per-image segmentation masks, enabling both illusion identification and localization benchmarks. It pairs data-driven models (ResNet for detection, UNet for localization) with diffusion-model based generation and validates the dataset via psychophysical experiments, achieving 99.56% identification accuracy and 84.37% pixel-level localization accuracy (mIoU 0.75). The work demonstrates generalization to unseen illusions and provides a pathway to study perceptual phenomena within computer vision, supported by public code and data. Overall, BRI3L offers a robust, psychophysically validated resource to advance understanding of visual illusions and their computational modeling.

Abstract

Visual illusions play a significant role in understanding visual perception. Current methods in understanding and evaluating visual illusions are mostly deterministic filtering based approach and they evaluate on a handful of visual illusions, and the conclusions therefore, are not generic. To this end, we generate a large-scale dataset of 22,366 images (BRI3L: BRightness Illusion Image dataset for Identification and Localization of illusory perception) of the five types of brightness illusions and benchmark the dataset using data-driven neural network based approaches. The dataset contains label information - (1) whether a particular image is illusory/nonillusory, (2) the segmentation mask of the illusory region of the image. Hence, both the classification and segmentation task can be evaluated using this dataset. We follow the standard psychophysical experiments involving human subjects to validate the dataset. To the best of our knowledge, this is the first attempt to develop a dataset of visual illusions and benchmark using data-driven approach for illusion classification and localization. We consider five well-studied types of brightness illusions: 1) Hermann grid, 2) Simultaneous Brightness Contrast, 3) White illusion, 4) Grid illusion, and 5) Induced Grating illusion. Benchmarking on the dataset achieves 99.56% accuracy in illusion identification and 84.37% pixel accuracy in illusion localization. The application of deep learning model, it is shown, also generalizes over unseen brightness illusions like brightness assimilation to contrast transitions. We also test the ability of state-of-theart diffusion models to generate brightness illusions. We have provided all the code, dataset, instructions etc in the github repo: https://github.com/aniket004/BRI3L

BRI3L: A Brightness Illusion Image Dataset for Identification and Localization of Regions of Illusory Perception

TL;DR

BRI3L introduces a large-scale, labeled dataset of 22,366 images capturing five brightness Illusions with per-image segmentation masks, enabling both illusion identification and localization benchmarks. It pairs data-driven models (ResNet for detection, UNet for localization) with diffusion-model based generation and validates the dataset via psychophysical experiments, achieving 99.56% identification accuracy and 84.37% pixel-level localization accuracy (mIoU 0.75). The work demonstrates generalization to unseen illusions and provides a pathway to study perceptual phenomena within computer vision, supported by public code and data. Overall, BRI3L offers a robust, psychophysically validated resource to advance understanding of visual illusions and their computational modeling.

Abstract

Visual illusions play a significant role in understanding visual perception. Current methods in understanding and evaluating visual illusions are mostly deterministic filtering based approach and they evaluate on a handful of visual illusions, and the conclusions therefore, are not generic. To this end, we generate a large-scale dataset of 22,366 images (BRI3L: BRightness Illusion Image dataset for Identification and Localization of illusory perception) of the five types of brightness illusions and benchmark the dataset using data-driven neural network based approaches. The dataset contains label information - (1) whether a particular image is illusory/nonillusory, (2) the segmentation mask of the illusory region of the image. Hence, both the classification and segmentation task can be evaluated using this dataset. We follow the standard psychophysical experiments involving human subjects to validate the dataset. To the best of our knowledge, this is the first attempt to develop a dataset of visual illusions and benchmark using data-driven approach for illusion classification and localization. We consider five well-studied types of brightness illusions: 1) Hermann grid, 2) Simultaneous Brightness Contrast, 3) White illusion, 4) Grid illusion, and 5) Induced Grating illusion. Benchmarking on the dataset achieves 99.56% accuracy in illusion identification and 84.37% pixel accuracy in illusion localization. The application of deep learning model, it is shown, also generalizes over unseen brightness illusions like brightness assimilation to contrast transitions. We also test the ability of state-of-theart diffusion models to generate brightness illusions. We have provided all the code, dataset, instructions etc in the github repo: https://github.com/aniket004/BRI3L
Paper Structure (24 sections, 28 figures, 6 tables)

This paper contains 24 sections, 28 figures, 6 tables.

Table of Contents

  1. Introduction
  2. Prior work
  3. BRI3L: A Brightness illusion image dataset
  4. Dataset generation
  5. Psychophysical experiment for dataset validation.
  6. Tasks: Illusion Identification, Localization and Generation
  7. Experiments
  8. Conclusion
  9. Supplementary Material
  10. Dataset curation details
  11. Psychophysical Experiment : Detail experimental setup, procedure and result.
  12. Background
  13. Materials and Methods
  14. Experimental Results
  15. Dataset
  16. ...and 9 more sections

Figures (28)

  • Figure 1: Examples of brightness illusion (top row) and the binary masks corresponding to certain illusory regions in the image (bottom row): (a) simultaneous brightness contrast (SBC), (b) White illusion, (c) Hermann grid, (d) grid illusion, (e) grating illusion. The illusory regions correspond to the illusory phenomenon such as apparent false perception of brightness/darkness values based on the context. Our goal is to localize these illusory regions in the illusory images.
  • Figure 2: (a): Actual Hermann grid; Weak or non-illusion variants of Hermann grid by (b) inserting blobs, (c) and (d) introducing non-linearity. geier2008straightness
  • Figure 3: Psychometric curve for (a)Lower grid, (b) SBC, (c) White illusion, (d) Grating illusion taken from the generated dataset. This indicates the probability that the perception agrees with the reality as a function of the real difference in luminance between the standard (S) and the comparator (C). The illusory reduction for Grating, White illusion, SBC, Lower grid are 26.69, 49.22, 35.03 and 32.95 respectively.
  • Figure 4: Different types of SBC illusions selected as the comparator. The intensity level of the target is always 150. However the length and width of the target varies widely.
  • Figure 5: Experimental setup for two-alternative forced-choice experiment for (a) SBC and (b) White illusion. Darker regions (red marked) are compared with the standard randomly multiple times to get the data.
  • ...and 23 more figures