Case study
Image classifier CIFAR-10
Convolutional neural network for image classification of the CIFAR-10 dataset
01 Task
The task was to train a model that classifies an image into one of ten CIFAR-10 classes, and to increase the accuracy to a reasonable level, and not stop at the baseline
02 Solution
Collected a convolutional network on Keras and TensorFlow, prepared and augmented the data, implemented convolutional blocks with regularization against overfitting, selection of hyperparameters. Conducted training and metrics in Jupyter, saved the final model for inference
03 Solved tasks
- Assembled a convolutional network using Keras and TensorFlow to classify 10 classes
- Increased accuracy above baseline by selecting hyperparameters
- Implemented convolutional blocks with regularization against overfitting
- Augmented the data and saved the trained model for inference
04 Result
The result is a working model with stable test accuracy and saved weights, which can be used for image recognition without retraining