deepapi 0.3.1

Deep API

Deep Learning as Cloud APIs.

This project provides pre-trained deep learning models as a cloud API service. A web interface is available as well.

• Quick Start
• API Client
  • Using curl
  • Using Python
• Concurrent clients
• Full APIs
  • Post URLs
  • Post Data (JSON)
  • Query Parameters
  • Returns (JSON)

Quick Start

Using Docker:

CPU:

docker run -p 8080:8080 wuhanstudio/deepapi

GPU:

docker run -p 8080:8080 wuhanstudio/deepapi:gpu

Python 3:

$ pip instlal deepapi
# By default, we enable all models on the server.
$ deepapi

Navigate to https://localhost:8080

API Client

It's possible to get predictions by sending a POST request to http://127.0.0.1:8080/vgg16_cifar10.

Using curl:

export IMAGE_FILE=test/cat.jpg
(echo -n '{"file": "'; base64 $IMAGE_FILE; echo '"}') | \
curl -H "Content-Type: application/json" \
     -d @- http://127.0.0.1:8080/vgg16_cifar10

Using Python:

def classification(url, file):
    # Load the input image and construct the payload for the request
    image = Image.open(file)
    buff = BytesIO()
    image.save(buff, format="JPEG")

    data = {'file': base64.b64encode(buff.getvalue()).decode("utf-8")}
    return requests.post(url, json=data).json()

res = classification('http://127.0.0.1:8080/vgg', 'cat.jpg')

This python script is available in the test folder. You should see prediction results by running python3 minimal.py:

cat            0.99804
deer           0.00156
truck          0.00012
airplane       0.00010
dog            0.00009
bird           0.00005
ship           0.00003
frog           0.00001
horse          0.00001
automobile     0.00001

Concurrent clients

Sending 5 concurrent requests to the api server:

$ python3 multi-client.py --num_workers 5 cat.jpg

You should see the result:

----- start -----
Sending requests
Sending requests
Sending requests
Sending requests
Sending requests
------ end ------
Concurrent Requests: 5
Total Runtime: 2.441638708114624

Full APIs

Post URLs:

Post Data (JSON):

{
  "file": ""
}

Query Parameters:

Example post urls (returns top 10 predictions with probabilities):

http://127.0.0.1:8080/vgg16?top=10&no-prob=0

Returns (JSON):

Example returned json:

{
  "success": true,
  "predictions": [
    {
      "label": "cat",
      "probability": 0.9996376037597656
    },
    {
      "label": "dog",
      "probability": 0.0002855948405340314
    },
    {
      "label": "deer",
      "probability": 0.000021985460989526473
    },
    {
      "label": "bird",
      "probability": 0.000021391952031990513
    },
    {
      "label": "horse",
      "probability": 0.000013297495570441242
    },
    {
      "label": "airplane",
      "probability": 0.000006046993803465739
    },
    {
      "label": "ship",
      "probability": 0.0000044226785576029215
    },
    {
      "label": "frog",
      "probability": 0.0000036349929359857924
    },
    {
      "label": "truck",
      "probability": 0.0000035354278224986047
    },
    {
      "label": "automobile",
      "probability": 0.000002384880417594104
    }
  ],
}

References

• https://github.com/geifmany/cifar-vgg

• https://keras.io/api/applications/