Results

Image Processing:

Receiver Operating Characteristics (ROC)

• The saliency maps include two different regions: salient and non-salient regions.
• Percentage of the salient regions from the ground-truth intersecting with the salient region from the saliency map is called the true positive rate.
• The ROC is also known as a relative operating characteristic curve because it is a comparison of two operating characteristics (TPR and FPR) as the criterion changes.

Saliency Maps

• In this project a database of 1000 images with their respective ground truths has been used.
• Images are used to find the saliency map and also to determine the performance characteristics of the received output i.e. the saliency map and its binary image.

Deep Learning:

Unsupervised Learning

• The result below is achieved using an Auto Encoder used for reducing dimensions to 64 from 4096.
• The Convergence was achieved at 624 Epoch with average cost(loss function) of 3.651632071e-05.
• Relu Activation function is used for introducing Non Linearity in the model to improve the efficiency of Model.