Implement of DeepLab ResNet training on Tensorflow

Content

1. Introduction
2. Dataset
3. Prerequisite
4. Tensorboard
5. Train
6. Evaluate
7. Test

Introduction

This implement folows the guide by DrSleep/tensorflow-deeplab-resnet and use transfer learning for a custome medical dataset. You can download the repo from here. All the files and settings are the same as above work other than specified in this document.

Dataset

There are two datasets employed to train the networks.

1. augmented PASCAL VOC 2012 dataset with 10582 images for training and 1449 images for validation. Images can be downloaded from here. Labels can be download from here. Following the structure in tensorflow-deeplab-resnet/dataset/train.txt to create folder and put in corresponding .png files. This dataset is used to validate that tensorflow works properly.
2. Retouch fluid retinal OCT dataset. The .raw files can be download from ReTouch. The .png files that can be directly used for training can be downloaded from here. This dataset is used to fine-tune the ResNet to segment medical content.

Prerequisite

1. you should have Tensorflow installed in your PC. Take a look of its version and make sure the DrSleep/tensorflow-deeplab-resnet supports your version. You can check the latest version it supports from Updates section of tensorflow-deeplab-resnet/README.md.
2. you shold clone the repo to your PC. You don't need to complete its instructoin at this point.

Tensorboard

To visualize training progress, model graphs, and internal state histograms: fire up Tensorboard and point it at your log_dir. E.g.:

$ tensorboard --logdir=./logs/

Then open a browser to http://localhost:6006 or the correct IP/Port specified.

Train

...under developing...