PyClassify

Python TensorFlow ResNet50 HTTP image classification server

This is a result of Google's TensorFlow team no longer supporting Go, a programming language made with their own (El Goog's) money.

Because I want to write my crawlers, which require image classification, in Go, but Python is the only Tensorflow supported language I really know, I have to do this somewhere, but my primary server for the site this is operating in service of doesn't even have enough ram to load tensorflow.

Run uing Docker

git clone https://github.com/somethingsoftware/PyClassify.git
cd PyClassify
docker build -t pyclassify .
docker run -dp 127.0.0.1:3000:3000 pyclassify

Run in Terminal

git clone https://github.com/somethingsoftware/PyClassify.git
cd PyClassify
pip install -r requirements.txt
python main.py

This is not built with any Python Web Framework, instead relying on the very basic http.server it expects a POST request to the endpoint /classify with a JSON-encoded JPEG image

{"image": "data:image/jpeg;base64,iVBORw0KGgoAAA..."}

where ... truncates a very long base-64 encoding of the image with or without the data:image/jpeg;base64,, that's just something my JS wanted to add.

it will respond with something like

{
    "success": true,
    "upload_size": 1,
    "classifications": {
        "television": "0.85180295",
        "desktop_computer": "0.09160688",
        "monitor": "0.04095296",
        "screen": "0.006704196",
        "notebook": "0.0039336607",
        "home_theater": "0.002433881",
        "iPod": "0.0011253094",
        "entertainment_center": "0.00035410884",
        "web_site": "0.00026599315",
        "desk": "0.00025780668"
    }
}

always the top 10 classifications and all their weights or some kind of error.