Skip to content

fvilmos/shape_color_position_detection

Folders and files

NameName
Last commit message
Last commit date

Latest commit

 

History

5 Commits
 
 
 
 
 
 
 
 
 
 

Repository files navigation

Multi headed DNN predictor, detects object coordinates, color and shape type

Toy DNN to demonstrate the usage of the multi-headed output in Keras. The task is to learn to detect basic shapes (circle, triangle, rectangle...and color) and the position of the shape center point. The target was to create a slim network with an optimal number of trainable variables ( this has ~290 k).

Steps:

  1. Create a toy database, generate shapes, color it. Prepare the data to be fed to DNN normalize values (values- mean)/ standard deviation and rescale the coordinate values from [0..128,0..128] -> [0..1,0..1].

  1. Create the multiheaded Keras functional model

  1. Train and test the DNN. After ~400 steps the network reaches about 93% accuracy on position prediction, 85% for shape, and 88% on colors. Take a look at the 'data' folder for better training results.

Links

  1. keras

/Enjoy.

About

A multi-headed DNN to detect shape type (circle, rectangle, triangle) shape color and position of the center point.

Topics

Resources

License

Stars

Watchers

Forks

Releases

No releases published

Packages

No packages published