Merge pull request #13 from medtune/stats_distribute

Enhance models training methods

Author: aelbouchti

.gitignore

@@ -8,6 +8,8 @@ __pycache__/
 
 # Distribution / packaging
 .Python
+env/
+research/
 build/
 develop-eggs/
 dist/
@@ -24,7 +26,7 @@ wheels/
 .installed.cfg
 *.egg
 MANIFEST
-
+*pyc
 # PyInstaller
 #  Usually these files are written by a python script from a template
 #  before PyInstaller builds the exe, so as to inject date/other infos into it.

DenseNet/preprocessing/densenet_pre.py

@@ -6,14 +6,14 @@
 
 slim = tf.contrib.slim
 #ImageNet RGB mean values (moyenne)0.485, 0.457,0.407
-_R_MEAN = 0.485
-_G_MEAN = 0.457
-_B_MEAN = 0.407
+_R_MEAN = 123.68/255
+_G_MEAN = 116.779/255
+_B_MEAN = 103.939/255
 
 #ImageNet standard deviation (écart-type)0.229, 0.224, 0.225
-_R_STD = 0.229
-_G_STD = 0.224
-_B_STD = 0.225
+_R_STD = 58.393/255
+_G_STD = 57.12/255
+_B_STD = 57.375/255
 
 _RESIZE_SIDE_MIN = 256
 _RESIZE_SIDE_MAX = 512
@@ -62,10 +62,6 @@ def _crop(image, offset_height, offset_width, crop_height, crop_width):
     tf.summary.image("rnd_cropped_image", tf.expand_dims(image,0))
     return tf.reshape(image, cropped_shape)
 
-
-
-
-
 def _random_crop(image_list, crop_height, crop_width):
 
     """Crops the given list of images.

README.md

@@ -1,14 +1,27 @@
 # Models-implementation
-This is an implementation of CNN's architectures using TensorFlow. The main goal of this project is to develop solutions for healthcare, based on machine learning & deep learning.
+This is an implementation of CNN's architectures using TensorFlow. The main goal of this project is to support the creation of machine learning & deep learning Pipelines .
+
+[lien](./setup.py)
 
 ## Description
 The repository is composed of:
 - Scripts for launching training and evaluation: It includes the basic tf.Session() and tf.train.MonitoredSession with different hooks and scaffold for distributed training.
 - Folder Densenet: Regroup image preprocessing functions, script for a slim-like implementation of Densenet
-- Folder utils: Convert data to Tfrecord format, analyse your data with statistics.py
+- Folder utils: Convert data to Tfrecord format, analyse your data depending on its nature
+
+##Installation
+
+First, install Tensorflow (Or/And tensorflow-gpu) in order to perform computation in cpu-only (In GPU) 
+Then, in order to use "slim" package, which is developped under : [lien](https://www.github.com/tensorflow/models/research/slim), 
+git clone the above repository. Copy the "research" folder and perform both setup.py : 
+- under research folder
+- under slim folder (make sure you delete BUILD file before running setup.py)
+
+As a last step, git clone Models-implementation repo : 
+https://www.github.com/medtune/Models-implementation
+
 
 ## Usage example
 
 
 ## Release History
-