Show and Tell: A Neural Image Caption Generator Demo API

Demo source files extracted from original TensorFlow Models source. (TensorFlow r0.10)

To build this demo, you don't need to prepare build environment with Bazel.

• Original repository: https://github.com./tensorflow/models/tree/master/im2txt
• Full text: http://arxiv.org/abs/1609.06647

Setup

• Tensorflow setup: https://www.tensorflow.org/versions/r0.11/get_started/os_setup.html
• Git Large File Storage setup: https://git-lfs.github.com./

git clone https://github.com./mainyaa/im2txt_api
cd im2txt_api
git lfs pull
pip install -r requirements.txt

cd im2txt/data
wget "http://download.tensorflow.org/models/inception_v3_2016_08_28.tar.gz"
tar -xvf "inception_v3_2016_08_28.tar.gz"
rm "inception_v3_2016_08_28.tar.gz"

Run simple

python -m im2txt.run_inference --input_files wii.jpg

Run API Server

sudo gunicorn im2txt.run_inference:app --log-file=-
open http://localhost:8000/

Hubot

script/im2txt.coffee

module.exports = (robot) ->
  robot.respond /im2txt +(.*)/i, (msg) ->
    # Send POST request
    request = require 'request'
    endpoint = 'http://[ADDRESS]/api/url?q=' + msg.match[1]
    request.get
      url: endpoint
    , (err, response, body) ->
      # Reply
      msg.reply "TensorFlow Show and Tell:\n```\n#{body}\n```"