A Comparison Between Human Visual Perception Under Object Segmentation and Recognition with State of the Art Deep Neural Networks.
- This work compares the attention of deep convolutional neural networks and the human visual perception system in classifying objects. In the proposed research, diagnostic regions for the human visual system and famous deep convolutional neural networks have been calculated; these regions are the most salient areas of each image, leading to accurate classification. They presumably have more meanings (than other regions) for each system, respectively.
- We computed the diagnostic features of each image in each category with five convolutional networks, (VGG16, ResNet50, EfficientNetb0, AlexNet and DenseNet-169); five saliency models (GBVS, Itti, Signature, Simpsal and Spectral); and finally, with the human visual perception system under a designed behavioral task.
- Have a look at following visual results for each section.
- On Deep CNNs
| VGG16 |  |
 |
 |
 |
 |
|---|---|---|---|---|---|
| ResNet-50 |  |
 |
 |
 |
 |
| DenseNet-169 |  |
 |
 |
 |
 |
| AlexNet |  |
 |
 |
 |
 |
| EfficientNet-b0 |  |
 |
 |
 |
 |
- On Saliency Models (will be added)
| GBVS |  |
 |
 |
 |
 |
|---|---|---|---|---|---|
| Itti-Koch |  |
 |
 |
 |
 |
| Signature |  |
 |
 |
 |
 |
| Simpsal |  |
 |
 |
 |
 |
| Spectral |  |
 |
 |
 |
 |
- You can also follow and check our project in Open Science Framework here.
- Understanding Vision 2021 Abstract and Presentation
- IICCSSS 2021 Abstract and Presentation
- ECVP 2021 Abstract, Youtube video, proposed Poster and Presentation
- Code explanation for permutation tests
- Behavioural task for obtaining diagnostic regions of human visual perception system is under construction and will be published soon.
If you had any feedback or question, please reach out to me at [email protected]
