Skip to content

Latest commit

 

History

History
40 lines (29 loc) · 1.55 KB

README.md

File metadata and controls

40 lines (29 loc) · 1.55 KB
Raw

Classification and object localization of the handwritten digits

Table of Contents

General Information

  • The project has two goals: 1) first to classifiy the handwritten digit 2) predict the bounding box around the hand-written digit
  • I am using a branched neural network architecture using TensorFlow functional API:
      1. feature_extractor: these convolutional layers extract the features of the image.
      1. classifier: This define the output layer that predicts among 10 categories (digits 0 through 9)
      1. bounding_box_regression: This defines the output layer that predicts 4 numeric values, which define the coordinates of the bounding box (xmin, ymin, xmax, ymax)

model

data

  • The model achieves an accuracy of 99% for the classification and MSE of 0.0012 for the regression on the validation set after 10 epochs.
  • The notebook has the capability to run on either TPUs, GPUs or CPUs. It uses tf.distribute.MirroredStrategy in case of multiple GPUs and tf.distribute.experimental.TPUStrategy in case of TPU.

Results

result

Technologies Used

  • Python
  • Tensorflow
  • Pandas
  • Matplotlib
  • Keras

Contact

Created by Miralireza Nabavi - feel free to contact me!