siddinc / resnet50

Custom ResNet implementation for Fruits-360 Dataset Image Classification

Image Classification using ResNet50

The goal of this project is to build and train a custom ResNet model with 50 layers which is able to classify different fruits and vegetables.

Tech used:

• TensorFlow 2.0.0
• Python 3.5.6

Dataset:

• Fruits-360 Dataset used for training and testing
• 82213 images of fruits and vegetables
• All images are RGB with dimensions 100 x 100 pixels
• Training set: 61488 images
• Test set: 20622 images
• Number of classes: 120

Images of each class are taken from all sides (360 degrees) of the fruit or vegetable.

Trained Models:

model1.h5 has the following accuracy metrics:

• Training accuracy = 99.21%
• Validation accuracy = 92.50%

model1.h5 was trained for 20 epochs with a batch size of 32

Instructions to run:

• Using anaconda:
  • Run conda create --name <env_name> --file tf2.yml
  • Run conda activate <env_name>
• Using pip:
  • Run pip install -r requirements.txt
• mkdir datasets in the same directory as src
• Download the Fruits-360 Dataset into datasets
• cd to src
• Run python main.py