zengsn / CNN_object_classification

Simple object classification project with deep-learning. We choose CIFAR10, CIFAR100 and Caltech101 as training datasets.

CNN_object_classification

Use CNN and VGG structure to train 3 datasets:

• CIFAR10
• CIFAR100
• Caltech101

Save the trained model, then use webcam and django to write a simple web application. link

Install

1. Install tensorflow and keras
2. Download caltech dataset
3. Run train/Train_CIFAR10_CIFAR100.ipynb or train/Train_caltech.py

Training Result

Dataset	Training acc	Testing acc	Runnung time (with GTX 960 8G)
CIFAR10	98%	80%	16 min
CIFAR100	89%	71%	7 hr
Caltech101	90%	65%	1 hr

Web application

The first block is the webcam preview window, and the little block near it is the image that we want to predict. After running backend prediction, it would show the predicted result.

Other result

CIFAR10
CIFAR100
Caltech101