CIFAR-10-Object Recognition in Images

Chenghao Ding, 26 February 2021

CIFAR-10-Object Recognition in Images

• Question: Identify the subject of 60,000 labeled images?

CIFAR-10 is an established computer-vision dataset used for object recognition. It is a subset of the 80 million tiny images dataset and consists of 60,000 32x32 color images containing one of 10 object classes, with 6000 images per class. The whole project can be found in my GitHub: (CIFAR10).

Data Description

The CIFAR-10 data consists of 60,000 32x32 color images in 10 classes, with 6000 images per class. There are 50,000 training images and 10,000 test images in the official data. There are 10 different lables in this dataset, such as: airplane, automobile, bird, cat, deer, dog, frog, horse, ship, truck. The first 36 images is shown above.

The datasets are separated into train( of size 50000), validation(10000), and test datasets(300000) respectively.

In this project, two commonly used CNN models: VGG and ResNet is implemented.

VGG

Pre-trained VGG16 and VGG19 are included in Keras, here, I build a VGG-like CNN models for object recognition.

Visualization of VGG model

Here is the screenshot of the output of model.summary().

VGG loss and accuracy versus training epochs

ResNet

The ResNet structure was re-implemented in Python.

ResNet loss and accuracy versus training epochs

The training/validation loss and accuracy versus training epochs are shown below. After 200 epochs, the test accuracy is around 0.90840. But looking carefully, after epoch 75, the training loss is still dropping while testing loss do not get any better since then. So the model seems to overfit.