sugyan/tensorflow-mnist

MNIST classification by TensorFlow

• MNIST For ML Beginners
• Deep MNIST for Experts

Requirement

• Python >=2.7 or >=3.4
  • TensorFlow >=1.0
• Node >=6.9

How to run

$ pip install -r requirements.txt
$ npm install
$ gunicorn main:app --log-file=-

Deploy to Heroku

$ heroku apps:create [NAME]
$ heroku buildpacks:add heroku/nodejs
$ heroku buildpacks:add heroku/python
$ git push heroku master

or Heroku Button.

tensorflow-mnist's Issues

bootstrap and jquery returns node_modules path

Hi,

I am trying to use your code to deploy mnist on heroku. I am not able to include jquery and bootstrap files. If I run, python main.py, everything runs fine but if I try to deploy using gunicorn or heroku, static/css/bootstrap.min.css returns node_modules path than the actual bootstrap file.

Example:
https://achal-mnist.herokuapp.com

not accurate in other digits such as ID cards's digits

hi,
i am so appreciated with your work. but i use your model to test for my digit cuted from ID-cards.
it seams not accurate.

Cannot mannually deploy to heroku

After creating the app and trying to add
$ heroku buildpacks:add heroku/nodejs
$ heroku buildpacks:add heroku/python
Both respond with app not found

Accuracy

I tried some digitals(such as 0, 5, 9, 7) on https://tensorflow-mnist.herokuapp.com/, it does not has a good performence(Both regression and tf are wrong). What can I do if I want to use it?

screencast

Deploy to Heroku : Build Error

Error:

$ pip install -r requirements.txt
       tensorflow-0.11.0-cp35-cp35m-linux_x86_64.whl is not a supported wheel on this platform.
 !     Push rejected, failed to compile Python app.
 !     Push failed

Screenshot:

The demo seem don't work well

Hi @sugyan

I go the the following message:
npm WARN cannot run in wd [email protected] gulp (wd=/home/srin/HandWritten/tensorflow-mnist)

and when I go to 127.0.0.1:8000, the demo don't work well. I

src under static/js and static/css may be wrong

Hi,
I think the src under static/js and static/css may be wrong.
I download these src from https://tensorflow-mnist.herokuapp.com/ manually.

Why the saved model in "data" folder is not .ckpt, but ckpt.meta?

I am new to tensorflow and I see that in the "data" folder it is like this after I delete the pretrained ckpt files and train it by myself:

So how to generate these pure .ckpt files?

Thanks!

Python and node?

Hi. Thanks for this awesome example!

I am wondering why you used both python and node.js?

Pre-built docker images are available now

I guess some people might prefer deploying their nodejs/flask app using docker images.
So I built some images here. Try:

docker pull liuqun/tensorflow-mnist:latest
# Or
docker pull liuqun/tensorflow-mnist:latest-fedora

(I found that the Fedora images is usually smaller than Debian based ones.)

Usage

docker run --rm -it -p 8000:8000 liuqun/tensorflow-mnist:latest
# Or
docker run --rm -it -p 8000:8000 liuqun/tensorflow-mnist:latest-fedora

The dockerfiles are here https://github.com/liuqun/tensorflow-mnist/tree/latest :

tensorflow-mnist/Dockerfile

Lines 1 to 27 in c1bd0fa

	# Dockerfile
	#
	# https://nodejs.org/en/docs/guides/nodejs-docker-webapp/
	FROM node:8.11.3-stretch
	RUN apt-get update \
	&& apt-get install -y --no-install-recommends \
	python2.7 \
	python-pip \
	python-setuptools \
	&& rm -rf /var/lib/apt/lists/*
	WORKDIR /usr/src/app
	COPY requirements.txt /usr/src/app/
	RUN pip install --no-cache-dir -r requirements.txt
	COPY package*.json gulpfile.js /usr/src/app/
	COPY src /usr/src/app/src
	RUN yarn install --ignore-engines
	COPY . /usr/src/app/
	EXPOSE 8000
	CMD [ "gunicorn", "main:app", "--log-file=-", "--bind=0.0.0.0:8000" ]

tensorflow-mnist/Dockerfile.fedora

Lines 1 to 30 in c1bd0fa

	# Dockerfile
	#
	# https://nodejs.org/en/docs/guides/nodejs-docker-webapp/
	FROM fedora:28
	RUN dnf install -y \
	libstdc++ \
	nodejs-1:8.11.0 \
	npm-1:5.6.0 \
	python3-3.6.5 \
	python3-pip \
	python3-setuptools \
	&& npm install --global yarn \
	&& dnf clean all
	WORKDIR /usr/src/app
	COPY requirements.txt /usr/src/app/
	RUN pip3 install --no-cache-dir -r requirements.txt
	COPY package*.json gulpfile.js /usr/src/app/
	COPY src /usr/src/app/src
	RUN yarn install --ignore-engines
	COPY . /usr/src/app/
	EXPOSE 8000
	CMD [ "gunicorn", "main:app", "--log-file=-", "--bind=0.0.0.0:8000" ]

git clone https://github.com/liuqun/tensorflow-mnist.git
cd tensorflow-mnist
git checkout latest
docker build -t mnist -f Dockerfile .
# Or: docker build -t mnist -f Dockerfile.fedora .

artifacts while resizing

I think, you have used following code to resize your canvas image, "(data[n + 0] + data[n + 1] + data[n + 2]) / 3" creates some artifact in the image. Is there easier way to just resize canvas image?

How do I send raw canvas image to flask? I am not an expert in javascript so tried some apis like getimagedata, base64encoding and everything but couldn't figure out. Thanks!

            var inputs = [];  
            var small = document.createElement('canvas').getContext('2d');  
            small.drawImage(img, 0, 0, img.width, img.height, 0, 0, 28, 28);  
            var data = small.getImageData(0, 0, 28, 28).data;  
            for (var i = 0; i < 28; i++) {. 
                for (var j = 0; j < 28; j++) {. 
                    var n = 4 * (i * 28 + j);  
                    inputs[i * 28 + j] = (data[n + 0] + data[n + 1] + data[n + 2]) / 3;  
                    ctx.fillStyle = 'rgb(' + [data[n + 0], data[n + 1], data[n + 2]].join(',') + ')';  
                    ctx.fillRect(j * 5, i * 5, 5, 5);  
                }. 
            }. 
            if (Math.min(...inputs) === 255) {. 
                return;  
            }.