EE211_finalproject

Include

EE211_finalproject
├── caffe
│   ├── get_train_log.sh
│   ├── get_test_log.sh
│   └── models
│  	│ 	│── bvlc_reference_caffenet
│	│	│	├── train_val.prototxt
│	│	│	└── solver.prototxt
├── ilsvrc12_train_lmdb
│   ├── data.mdb
│   └── lock.mdb
├── ilsvrc12_val_lmdb
│   ├── data.mdb
│   └── lock.mdb
├── imagenet_mean.binaryproto
├── create_imagenet.sh
├── make_imagenet_mean.sh
├── train_test.sh
├── train.txt
└── val.txt

Introduction

This project is Caffenet-based Automatic Prostate Tumor Classification from T2-MRI. The framework is caffe and we build our network based on BVLC reference caffenet and set our own solver file. We use PROSTATEx Challenge Data Sets for training and testing and random pick-in, random pick-out to augmentation the image. The final accuracy can get an average value of 0.7.

Dependencies

1. Download data from our google drive which included in data.md.
2. Set up caffe environment and the instruction is included in mac_caffe.md.
3. Use our train_val.prototxt and solver.prototxt to replace the original files in models/bvlc_reference_caffenet.
4. Prepare data and use our script file to train model and get the result.

Data preparation

1. Download data and set the fold name as train and val.
2. Put train and val image fold and your caffe framework fold in your work space fold.
3. Run bash train_test.sh train will create train.txt to save the train image path and its label in a txt file.
4. Run bash train_test.sh val will create val.txt in the same way.
5. Run ./create_imagenet.sh will create train and val lmdb file which will help to increase the calculate efficiency.
6. Run ./make_imagenet_mean.sh will create imagenet_mean.binaryproto which is useful in training.

Training model and save the result

1. Make sure you have get train and val lmdb file, imagenet_mean.binaryproto file and you have build your caffe successfuly.
2. Make sure you have successfully replced train_val.prototxt and solver.prototxt as the Dependencies part.
3. Get into your caffe fold and run get_train_log.sh bvlc_reference_caffenet to apply adjusted training file. If you want to test other network. You just need to rewrite your training file and change the model name to yours.
4. After training, you will find train_result.txt which saves the training log in caffe_train_log fold.
5. If you want to test the model, you can run get_test_log.sh bvlc_reference_caffenet which saves the predict log in caffe_net_log fold.

Sample

We save some logs of our training model in training_log fold as a sample.