first commit

Author: aurelienmorgan

.gitattributes

@@ -0,0 +1,12 @@
+
+# exclude from GitHub's programming-language statistics
+/**/*.html linguist-vendored
+/**/*.ipynb linguist-vendored
+
+# These files are text and should be normalized (Convert crlf => lf)
+*.gitattributes text
+.gitignore      text
+*.md            text
+
+# Enable syntax highlighting for files with `.gitattributes` extensions.
+*.gitattributes linguist-language=gitattributes

.gitignore

@@ -0,0 +1,16 @@
+
+.vs/
+
+images/web_app.xcf
+flask_app/__pycache__/
+flask_app/app/__pycache__/
+flask_app/app/media-offline/*
+flask_app/app/static/images/banner.xcf
+flask_app/app/static/images/docker_network.xcf
+flask_app/app/static/images/target_blank.xcf
+flask_app/app/static/images/target_blank.png
+flask_app/*.proj
+flask_api/__pycache__/
+flask_api/api/utils/__pycache__/
+docker_archives/
+README.txt

README.md

@@ -0,0 +1,61 @@
+# Surface Defect Detector
+
+Welcome to this demonstration platform  ! 
+
+The goal is to demonstrate how to easily integrate a Tensorflow/Keras model into a microservice architecture to provide predictions on the fly.
+
+The use case is surface defect detection from flat steel sheet images.
+
+The architecture employed is made-up of 3 Docker containers as follows : 
+
+![](flask_app/app/static/images/docker_network.png)
+
+The key element here is the webservice named ```defect-api-service```
+which is responsible for the generation of the prediction.
+Given an input image, it operates the surface defect detection
+and returns an augmented image and probability informations.
+
+To illustrate its usage, the prediction webservice has been integrated into
+a basic web platform :
+
+>```http://localhost/upload``` :
+select the image for anaysis<br />
+<img alt="upload page" src="images/upload.png" width="650px" />
+
+<br />
+
+>```http://localhost/prediction?filename=5e1c6b7da.jpg``` :
+result of the prediction as returned by ```defect-api-service```<br />
+<img alt="prediction page" src="images/prediction.png" width="650px" />
+
+<br />
+
+>```http://localhost/media/``` :
+list of media file<br />
+<img alt="media page" src="images/media.png" width="650px" />
+
+For each image submitted for defect detection,
+the ```defect-api-service``` stores three different files :
+- the original image itself
+- a json representation of the prediction
+- the image with an overlayed contour
+of the detected defectuous area (if any)
+
+<br />
+
+For details on the model architecture and training
+as well as on what the reported probabilities do measure,
+please kindly refer to <a alt="jupyter notebook" href="https://htmlpreview.github.io/?https://github.com/aurelienmorgan/defect_detection_webservice/blob/master/notebook/model.html?uncache=654645">this walkthough Jupyter Notebook</a>.
+
+<br />
+
+
+
+
+KEYWORDS :
+	```Nginx```, ```uwsgi```, ```Gunicorn```,
+	```container```, ```Docker```,
+	```Flask```, ```REST api```, ```webservice```,
+	```Tensorflow```, ```Keras```, ```U-Net```,
+	```segmentation model```, ```data augmentation```,
+	```OpenCV```

docker-compose.yml

@@ -0,0 +1,60 @@
+version: "3.7"
+
+services:
+
+    defect-api-service:
+        # build the container, looking into the Dockerfile at the bellow relative path
+        build: ./flask_api
+        container_name: flask_api
+        image: flask_api:v0-phil
+        restart: always
+        # A place for us to define environment variables for the container
+        environment:
+            - PYTHONUNBUFFERED=1 # stdout not buffered
+            - APP_NAME=MyFlaskApi
+            - MEDIA_FOLDER=/api/api/media
+        volumes:
+            - media_volume:/api/api/media:rw
+        # Exposes internal ports to other containers and services on the same network
+        expose:
+            - 9090
+
+    web_app:
+        # build the container, looking into the Dockerfile at the bellow relative path
+        build: ./flask_app
+        container_name: flask_app
+        image: flask_app:v0-p
+        restart: always
+        # A place for us to define environment variables for the container
+        environment:
+            - APP_NAME=MyFlaskApp
+            - STATIC_FOLDER=/app/app/static
+            - MEDIA_FOLDER=/app/app/media
+        volumes:
+            - static_volume:/app/app/static:ro
+            - media_volume:/app/app/media:rw
+        # Exposes internal ports to other containers and services on the same network
+        expose:
+            - 8080
+        # declare service dependencies
+        depends_on:
+            - defect-api-service
+
+    http_server:
+        build: ./http_server
+        container_name: http_server
+        image: http_server:v0-guk
+        restart: always
+        volumes:
+            - static_volume:/home/web_app/static:ro
+            - media_volume:/home/web_app/media:ro
+        # expose ports to the outside world,
+        # mappes port 80 on the host machine to port 80 of our Nginx container.
+        ports:
+            - "80:2020"
+        depends_on:
+            - web_app
+
+volumes:
+    static_volume:
+    media_volume:

flask_api/.dockerignore

@@ -0,0 +1,2 @@
+api/__pycache__/
+api/utils/__pycache__/

flask_api/Dockerfile

@@ -0,0 +1,25 @@
+# Uthe Python 3.7.2 container image
+FROM python:3.7.2-stretch
+
+# Set the working directory to "api"
+WORKDIR /api
+
+# Copy the current directory contents into the container at /api 
+ADD . /api
+
+# Install the dependencies
+RUN pip install -r requirements.txt
+
+
+
+# create an empty "/api/api/media/" folder prior to (volume creation &) parent permission allocation
+RUN mkdir -p /api/api/media && \
+	chown -R www-data:www-data /api && \
+	chown -R www-data:www-data /dev/shm
+#RUN stat /api
+#RUN cd api && stat media && cd ..
+
+USER www-data
+
+# Command to start Gunicorn
+ENTRYPOINT ["./gunicorn_starter.sh"]

flask_api/api/__init__.py

@@ -0,0 +1,6 @@
+
+import os
+
+from flask import Flask
+
+api = Flask(__name__)