# Visual Illusion Infos
# 01-18 Color Dataset @ Jiayi Pan
Note:
- 6: TODO
- 9: wite edge
- 12: no "background remove group"

## Naming Convention
Type_Idx_Variant?

Variant:
- N/A: standard illusion image
- r: remove background
- a: actually have difference (not maintained well)

# Previous Note

## Dataset TODO
1. 第一是铁轨那种基于透视的illusion,不能颠倒，不然没illusion
2. lines are too short
3. remove background
4. slight color augmentation (4 samples would definitely be enough)
5. paper画图

# Related Works
> This section only includes related works about the computationaly models. How these illusions are built/why they exists are included in the dataset section

- [Convolutional Neural Networks Can Be Deceived by Visual Illusions](https://openaccess.thecvf.com/content_CVPR_2019/html/Gomez-Villa_Convolutional_Neural_Networks_Can_Be_Deceived_by_Visual_Illusions_CVPR_2019_paper.html)
- [ImageNet-trained deep neural networks exhibit illusion-like response to the Scintillating grid](https://jov.arvojournals.org/article.aspx?articleid=2778014)
- [Color Visual Illusions: A Statistics-based Computational Model](https://arxiv.org/abs/2005.08772#:~:text=Visual%20illusions%20may%20be%20explained,to%20extensively%20support%20these%20explanations.)
- [Optical Illusions Images Dataset](https://arxiv.org/abs/1810.00415)
- Unimodal, image with sparse meta-data
- [A comparative biology approach to DNN modeling of vision: A focus on differences, not similarities](https://jov.arvojournals.org/article.aspx?articleid=2777922)
- "we argue that the main problem is that we often do not understand which human functions need to be modeled and, thus, what counts as a falsification. Hence, not only is there a problem on the DNN side, but there is also one on the brain side (i.e., with the explanandum—the thing to be explained). For example, **should DNNs reproduce illusions?**"
- [Rethinking the Maturity of Artificial Intelligence in Safety-Critical Settings](https://ojs.aaai.org/index.php/aimagazine/article/view/7394)
- has some

# Dataset
Stats:
1. Base image: 14
2. Augmented total images:
3. Images for ablation:
## Color
###### Reference
[A Brief Classification of Colour Illusions](https://hackmd.io/@ERAt5jzHTIitAd5c2nB03A/SJleH-mso)
### Assimilation
#### Hue
##### Hue-1

Augmentation
1. Change background/cross color: 02_h01 - 02_h06
2. Bottom-up / left-right configuration

Ablation:
1. Remove background: : 02_r01 - 02_r06
2. Make the cross actually has different colors: : 02_h01_d - 02_h06_d

##### Hue-2

Augmentation
1. Change background/ball/line color: 01_h01 - 01_h06
2. Bottom-up / left-right configuration

Ablation:
1. Remove background/lines: 01_r01 - 01_r06
2. Make the ball actually has different colors: 01_r01_d - 01_r07_d

#### Brightness
##### Brightness-1

Augmentation
1. Change background/ball/line color: 02_b01 - 02_b07
2. Bottom-up / left-right configuration

Ablation:
1. Remove background/lines: 02_r11 - 02_r17
2. Make the cross actually has different colors: 02_b01_d - 02_b07_d

##### Brightness-2
Augmentation
1. Change background/ball/line color: 01_b01 - 01_b16
2. Bottom-up / left-right configuration
3. Ball position: 01_b21 - 01_b26

Ablation:
1. Remove background/lines: 01_r01 - 01_r06
2. Make the cross actually has different colors: 01_b31_d - 01_b36_d
### Contrast
- TODO
##### Contrast-1

Augmentation
1. Change background: 03_01 - 03_06
2. Bottom-up / left-right configuration

Ablation:
1. Make the two blocks actually have different colors: 03_01_d - 03_06_d

##### Contrast-2

Augmentation
1. Change background/ball color: 04_01 - 04_07
2. Bottom-up / left-right configuration

Ablation:
1. Remove background: 04_r01 - 04_r07
2. Make the two balls actually has different colors: 04_01_d - 04_07_d

### Constancy

Augmentation
1. Change filter color / object in the image: 06_01 - 06_02, 07_01
2. Bottom-up / left-right configuration

Ablation:
1. Remove the filter: 06_r01 - 06_r03, 07_r01
2. Make the two objects actually has different RGB colors: 06_01_d - 06_02_d, 07_01_d

###### References
https://www.researchgate.net/publication/228490244_A_Brief_Classification_of_Colour_Illusions

###### Explanation
Colour constancy refers to a phenomenon that observers can see the “true” colour of an object to some extent even if illumination is changed in colour.

### Cognitive(TODO) (1)

Ablation:
1. Make the labeld squares actually has different RGB colors: 05_d

###### Reference
http://persci.mit.edu/gallery/checkershadow

###### Explanation:

The visual system needs to determine the color of objects in the world. In this case the problem is to determine the gray shade of the checks on the floor. Just measuring the light coming from a surface (the luminance) is not enough: a cast shadow will dim a surface, so that a white surface in shadow may be reflecting less light than a black surface in full light. The visual system uses several tricks to determine where the shadows are and how to compensate for them, in order to determine the shade of gray “paint” that belongs to the surface.

The first trick is based on local contrast. In shadow or not, a check that is lighter than its neighboring checks is probably lighter than average, and vice versa. In the figure, the light check in shadow is surrounded by darker checks. Thus, even though the check is physically dark, it is light when compared to its neighbors. The dark checks outside the shadow, conversely, are surrounded by lighter checks, so they look dark by comparison.

A second trick is based on the fact that shadows often have soft edges, while paint boundaries (like the checks) often have sharp edges. The visual system tends to ignore gradual changes in light level, so that it can determine the color of the surfaces without being misled by shadows. In this figure, the shadow looks like a shadow, both because it is fuzzy and because the shadow casting object is visible.

The “paintness” of the checks is aided by the form of the “X-junctions” formed by 4 abutting checks. This type of junction is usually a signal that all the edges should be interpreted as changes in surface color rather than in terms of shadows or lighting.

As with many so-called illusions, this effect really demonstrates the success rather than the failure of the visual system. The visual system is not very good at being a physical light meter, but that is not its purpose. The important task is to break the image information down into meaningful components, and thereby perceive the nature of the objects in view.

## Size
### Relativity

Augmentation
1. Change color / object in the image
2. Bottom-up / left-right configuration

Ablation:
1. Remove the outer objects
2. Make the two inner objects actually has different size

### Perspective (4 / )
TODO: is it known as "geometric illusion"?
- ponzo illusion
- Muller Lyer illusion

Augmentation
1. Change perspective / object in the image
2. Bottom-up / left-right configuration when possible

Ablation:
1. Remove the background
2. Make the two objects actually has different size

## Symbol (1 / )

###### References
https://www.tandfonline.com/doi/abs/10.1080/00221309.1955.9710133

###### Explnation
Basicially demonstrate the "Influences of spatial context on dynamic perceptually unstable figures."
https://search.proquest.com/openview/628eb79c25ac49715344132545d6cc74/1?pq-origsite=gscholar&cbl=18750&diss=y

TODO: we might want to change the category name

# Left out

## COLOUR ILLUSION BY MOTION
Our probed model is static

## AFTERIMAGE
## Paradigm