In this project, we use flickr style dataset to demonstrate finetune in TensorFlow. The details please refer to the example from the Caffe website
Thank @ethereon for his code. We modify the network.py from caffe-tensorflow for our use.
# Download dataset in the current folder
$ python assemble_data.py
Please follow the tutorial and extract caffenet.npy
https://github.com/ethereon/caffe-tensorflow
# Dataset path
train_list = '/path/to/data/flickr_style/train.txt'
test_list = '/path/to/data/flickr_style/test.txt'
$ python finetune.py
// Fine-tuning result
Iter 640: Testing Accuracy = 0.2801
// From scratch result
Iter 640: Testing Accuracy = 0.1588