This resource currently contains hundreds of facial images (AI created, real photos, cartoons etc.) with varied emotions (with royalty free usage permit). In addition, the likely “diagnostic efficiency” of the images (i.e., potential ability to optimally discriminate between behaviors of individuals with autism and neurotypical development) as predicted by various models of primate vision are also provided here. This is a work in progress and I hope over time, the research community around this will grow to include folks doing clinical/translational work who can test and validate these predictions
All the functionalities of the website will be soon enabled
Basic Steps:
Original Image | Resized and SHINED | Aperture placed | Pass through ANN | Predicted Difference |
---|---|---|---|---|
0.13 |
Check-out the image gallery:
Background: Why does this webpage exist ??
What is DECIFER Autism?
Main Goal
Scientific Motivation
Working hypothesis and preliminary proof of principle
Current available images and corresponding model predictions
This will grow over time
Model Name | Full-set DE score | Optim-set DE score | Download (All) | Download (optimized) |
---|---|---|---|---|
AlexNet. | 0.34 | 0.84 | Link (activated upon publication) | Link(activated upon publication) |
Emotion Net | 0.25 | 0.94 | Link (activated upon publication) | Link(activated upon publication) |
VGG-Face | 0.32 | 0.88 | Link(activated upon publication) | Link(activated upon publication) |
CORnet-S | 0.15 | 0.83 | Link(activated upon publication) | Link(activated upon publication) |
Future additions coming!
* Morphs between facial emotions of same individuals
* Videos of individuals demonstrating facial emotions
* Predictions from more models
* Neurally regularized model predictions
* Model synthesized images with specific behavioral predictions
Support or Contact
If you have any questions, advice, suggestions, eagerness to collaborate, or want to discuss the science – related to this line of research, please email me: Kohitij Kar (you can refer to me as Ko if that is easier)