BuildingAI - My AI idea

This is the final project for the Building AI course.

Summary

My AI idea will be an image classifier. It can be done with TensorFlow, a python library used for machine learning. The idea is to simulate a shop and the program will classify the photos we pass by their shape.

Background

This idea can solve the bad classification of the products of a shop, sometimes the people can make mistakes, and I think that having a classifier would solve some of them. I think this project can be a good way to start practicing with some of the concept I learnt in this course.

The idea is to compare the pixels of the images to give the AI an idea of what kind of items appears on them.

How is it used?

This program will be used, of course, when new stock arrives to the shop. The employee will only have to make a photo to every item and put them in the program, and them will be classified automatically.

One example of a function that will be in the program is the next one, here we have an image resizer. To improve the pixel detection, all the images will be shrinked to 100x100px:

def graficar_imagen(i, arr_predicciones, etiquetas_reales, imagenes):
  arr_predicciones, etiqueta_real, img = arr_predicciones[i], etiquetas_reales[i], imagenes[i]
  plt.grid(False)
  plt.xticks([])
  plt.yticks([])
  
  plt.imshow(img[...,0], cmap=plt.cm.binary)

  etiqueta_prediccion = np.argmax(arr_predicciones)
  if etiqueta_prediccion == etiqueta_real:
    color = 'blue'
  else:
    color = 'red'
  
  plt.xlabel("{} {:2.0f}% ({})".format(nombres_clases[etiqueta_prediccion],
                                100*np.max(arr_predicciones),
                                nombres_clases[etiqueta_real]),
                                color=color)

Data sources and AI methods

The program will work with Neural Networks, which will be programmed, as I said, using the open source python library TensorFlow.

Challenges

The photos have to be taken on a simple background, the main object must be easily identified, it is a simple classifier.

What next?

My AI knowledge cannot think any other new implementations to the program.

Acknowledgments