InceptionV3_TensorFlow is an implementation of inception v3 using tensorflow and slim according to our guidline.
- TensorFlow (>= 0.12)
- train
- predict
- save checkpoint
- real time data augumentation
If you want a quick start to run training of Inception_v3, you can simply do:
./train.sh
The above script has passed test under Ubuntu15.10, CentOS and macOS.
If you want to go through the train process step by step, please take the following content as example.
- download data in data/readme.md
- execute "data/create_examples_list.py"
- execute "data/relation_tag_to_id.py"
- you can see train_csv.txt and test_csv.txt
python trainer.py
Pass test under Ubuntu15.10 and CentOS
- create train_csv.txt and test_csv.txt in data directory.
<image path>,<label number>
...
- change num_classes in settings.py
- change fine_tune in settings.py
tensorboard --logdir /to/your/path/train_dir --port=6006
Copyright (c) 2016 Masahiro Imai, Yixuan Hu (yeephycho) Released under the MIT license