https://github.com/furk4neg3/environment-image-classification

Image classification model using intel image classification dataset. Created 6 models using various techniques, which involves CNN and fine-tuning. Model works with 87% accuracy.
https://github.com/furk4neg3/environment-image-classification

artificial-intelligence cnn-classification deep-learning fine-tuning machine-learning tensorflow tensorflow2

Last synced: about 1 month ago
JSON representation

Image classification model using intel image classification dataset. Created 6 models using various techniques, which involves CNN and fine-tuning. Model works with 87% accuracy.

• Host: GitHub
• URL: https://github.com/furk4neg3/environment-image-classification
• Owner: furk4neg3
• Created: 2024-08-10T22:00:11.000Z (4 months ago)
• Default Branch: main
• Last Pushed: 2024-08-10T22:04:48.000Z (4 months ago)
• Last Synced: 2024-08-10T23:19:00.730Z (4 months ago)
• Topics: artificial-intelligence, cnn-classification, deep-learning, fine-tuning, machine-learning, tensorflow, tensorflow2
• Language: Jupyter Notebook
• Homepage:
• Size: 2.41 MB
• Stars: 1
• Watchers: 1
• Forks: 0
• Open Issues: 0
• Metadata Files:
  • Readme: README.md

Awesome Lists containing this project

README

        
# Environment Image Classification

In this notebook, I've used Intel's image classification dataset. This involves images of scenes around the world. I've created 6 AI models for this problem. I've made data augmentation and preprocessing too.

-> First model is a baseline model which doesn't use deep learning.

-> Second model is a small dense model, which acts like baseline for deep learning models.

-> Third model is a small convolutional model, which only involves convolutional layers.

-> Fourth one is a bigger convolutional model, which involves max pooling layers.

-> Fifth one is the biggest of my self-created models, which is also a convolutional model. This one has more convolutional layers, pooling layers and involves batch normalization layers as well.

-> Last model uses fine-tuning, transfer-learning model is Inception v3 (I've created this model's v1 in TensorFlow, it's also in my GitHub).

-> Every model's loss and accuracies are visualized while training, and at the end they've been compared visually.

-> Best model which has 87% accuracy has been chosen, then I made a prediction with that model on a photo that my girlfriend sent me too, and it predicted the right label.