chencgln / Finetune-Inception-Restnet-V2

An example of fine tuning inception_resnet_v2 using TF-slim. Get prediction for single input image.

Finetune-Inception-Restnet-V2

An example of fine tuning inception_resnet_v2 using TF-slim

Prerequisite

This project is built and tested with tensorflow 1.8.

please make sure the following items are ready on your machine.

1. TensorFlow
2. TF-slim: make sure path is added to $PYTHONPATH
3. Run scripts under TensorFlow env

Scripts

This Repository provide scripts to quickly run training and evaluation. You can also change parameters in those scripts or write your own script, to train on other datasets or models.

The model is trained under GPU mode by default. You can use CPU mode by setting the FLAG --clone_on_cpu=True for python script: train_image_classifier.py.

Run these script in the root path of this project.

prepare_dataset&ckpt.sh:

Download the flowers dataset and checkpoint file inception_resnet_v2_2016_08_30.tar.gz

train_inception_resnet_v2_classifier.sh:

Feed in the flowers dataset and train the pre-trained inception_resnet_v2. The checkpoint file will be saved in to train/flowers/inception_resnet_v2 by default.

evaluate_classifier.sh:

Evaluate the performance of the trained model.

predict_single_image.sh:

Feed in a single image in to the model and get the prediction results.