Transfer-learning-with-ResNet-50-in-Keras

Introdution

This project is aiming to train a image classification model by transfer learning with ResNet50 pre-trained model. It is impelemented by Keras.

Requirements

conda for open source package management and environment management

h5py==2.10.0

matplotlib

keras==2.2.5

tensorflow==1.14

Usage

Step1: Install conda

visit conda installation & install the compatible version on your system environment.

Step 2: Create virtual environment for this project

conda create -n tl-resnet50 keras==2.2.5

Step 3: Activate this env

conda activate tl-resnet50

Step 4: Install requirements

pip install -r requirements

Step 5: Create folders for storing data & models

mkdir data && mkdir models

Step 6: Put your dataset into data folder

Your folder structure should be like the following format

data > 
  your_dataset > 
    train > 
      class_1 > 
        img_1.jpg
        img_2.jpg
        img_3.jpg
        ...
      class_2 >
        img_1.jpg
        img_2.jpg
        img_3.jpg
        ...
    validation >
      class_1 > 
        img_1.jpg
        img_2.jpg
        img_3.jpg
        ...
      class_2 >
        img_1.jpg
        img_2.jpg
        img_3.jpg
        ...

Step 7: Replace 'input_path' value in train.py & predict.py

in both file

input_path = './data/your_dataset/'

in predict.py

validation_img_paths = ['./validation/img_1.jpg', './validation/img_2.jpg', './validation/img_3.jpg']

Step 8: Train model & Validate training model

Training model:

python train.py

Validate training result:

python predict.py

lbj96347/Transfer-learning-with-ResNet-50-in-Keras