Classifier From Multple S Guide

User Manual:

Open the PDF directly: View PDF PDF.
Page Count: 5

DownloadClassifier From Multple S Guide
Open PDF In BrowserView PDF
Date: November 28, 2018

MODEL TRAINING GUIDE

1. Deep learning Libraries setup
Step 1: Install Python
You can download the latest version of Python from here :
https://www.python.org/downloads/
Step 2: Download Anaconda
https://www.anaconda.com/download/
Step 3: Install OpenCV
pip install opencv-python
All the libraries that we are using will be covered under this and included
under conda. Even if at the time of executing code, it demands installation
of any package, it could be easily installed using command" ​pip install 'package-name' ​".
2.Environment setup
You can create an environment which is like an isolated virtual container
where you can install python 3.6 and install your libraries.

conda list envs — Lists all the environments created in Anaconda
conda activate env_name — To activate and start working in a particular
environment
I have activated "tensorFlow_p36" for our project.
Command for that would be source activate tensorflow_p36​ , or
conda activate tensorflow
3. Folder ArchitectureInitially our directory will look like:
Classification-Model-train_test_split.py
-Data_prepration.py
-train.py
-test.py
-positive_paths.txt (Contains paths of folders having positive images)
-negative_paths.txt (Contains path of folders having negative images)
Later after compiling python files in sequence,some files would
automatically get created. At the end our folder would look likeClassification-Model
-train_test_split.py
-positive_training.txt
-positive_test.txt
-negative_training.txt
-negative_test.txt
-Data_prepration.py
-train.py
-test.py
-Accuracy.jpg (Graph that shows accuracies of model while training)

-X_concrete.pickle
-y_concrete.pickle
-Concrete_Crack_Classification_model.model (our model after
creation)
-logs.json (it's a json file containing model history)
-positive_paths.txt
-negative_paths.txt
-Residuals.txt (Contains paths for images which are detected False
positive, False negatives and those who couldn’t be complied due to
some error.)
4.Running Model
Step 1:In Pipeline, there should be two text files. Each of them containing
corresponding paths for folders containing Positive and Negative Images.
Step 2: In terminal, traverse into the Pipeline folder.
Step 3: Activated the Environment as explained in point-2
ex. source activate tensorflow_p36
Now we just had to run this four files in given order

-train_test_split.py --pos= *positive txt file* --neg= *negative txt file*
-Data_prepration.py
-train.py
-test.py
Step 5: Splitting data into train and test
terminal command :
python train_test_split.py --pos=*positive txt file* --neg=*negative txt file*
example :
python train_test_split.py --pos=pospath.txt --neg=negpath.txt

It will create 4 text files for positive and negative images for both training
and testing respectively.
-positive_training.txt (Contains paths of positive images for Training)
-positive_test.txt (Contains paths of positive images for Testing)
-negative_training.txt (Contains paths of negative images for
Training)
-negative_test.txt (Contains paths of negative images for Testing)
Note - Currently splitting into 80% train data and 20% test data ,it could be
changed accordingly.
Step 6: Data Preprocessingterminal command : python Data_prepration.py
It will create two pickle files from training data.
-X_concrete.pickle
-y_concrete.pickle
Step 7: Initiate Training
terminal command : python train.py
It will start training our model, and generate an ‘.jpg’ image file displaying
accuracy graph.
-Accuracy.jpg
Note- Currently Epochs are set to 10 and validation split to 20 percent, it
could be changed accordingly.
Step 8: Testing on sample
terminal command : python test.py
It will show the inferences, Accuracy of our model at the end. It will also
create a text file “Residuals.txt” which would contain paths for images
which are predicted false positive, false negative and images which gave
error while compiling.
- Residuals.txt

All Done!!
Source Exif Data: 
File Type                       : PDF
File Type Extension             : pdf
MIME Type                       : application/pdf
PDF Version                     : 1.5
Linearized                      : Yes
Producer                        : Skia/PDF m73
Page Count                      : 5
EXIF Metadata provided by EXIF.tools

Navigation menu