TensorFlow ResNet50

Custom implementation of ResNet50 Image Classification model using pure TensorFlow

Requirements

Python 3.7
Tensorflow 1.x

Dataset Requirements

Dataset Folder should only have folders of each class. Dataloader will automatically split the dataset into training and validation data in 80:20 ratio.
Example:

.
└── DatasetFolder
    ├── ClassOne                 
    │   ├── FirstImage.jpg                       
    │   ├── SecondImage.jpg                 
    │   └── ...    
    ├── ClassTwo  
    │   └── ...    
    ├── ClassThree               
    │   └── ...    
    └── ...

Usage

Training

python train.py -e=[number of epochs] -f=[dataset folder path] -d=[optional: if use TF Debugger]

TensorBoard

To see metrics while training, run tensorboard.
Plotted metrics are:

Each batch accuracy, both train and val
Each batch loss, both train and val
Epoch accuracy, both train and val
Epoch loss, both train and val

tensorboard --logdir=logs

Prediction

python predict.py -img=[path to fodler with images awaiting prediction] -f=[path to dataset folder] 
-mod=[path to saved model folder] -d=[optional: if use TFDebugger]

Project Structure

.
├── data                       
│   ├── data.py                 # Dataloader  
│   └── utils.py                # Image Parser
├── model                       
│   ├── resnet.py               # Resnet50 Model
│   └── layers.py               # Model's Layers 
├── logs                        # TensorBoard Logs         
├── training                    # Model's Weights
├── config.json                 # Configuration File
├── train.py                    # Training Script
└── predict.py                  # Preidction Script

sevakon/resnet50-tensorflow