Deep learning to detect objects from an image.

The code and details are provided in the repository shared

The code applies advanced deep-learning techniques to detect the foreground and background within the source image.

Advanced statistical techniques including entropy loss, focal loss and partial cross-entropy loss implementation can be seen within the documentation.

Heat maps using the Logits and Seaborn are automatically created once you run the code.

The Loss Function

The Results

Some samples for character detection. The color represents the probability of each pixel being the subject. The brighter the color the higher the probability.