ceevee

ceevee (read like CV, i.e. computer vision) is a Python library for various computer vision problems with a focus on easy usage.

ceevee aims to be a bridge between deep learning practitioners training accurate models and product-oriented software engineers who just want to process their images instead of diving into the deep learning ecosystem.

Python 3.6+ is supported.

Install

From PyPI - not available yet

From source

python setup.py bdist_wheel
pip install -U ceevee-0.0.1-py3-none-any.whl

Tasks

Usage

All tasks shares the same API

CLI API

python -m ceevee.cli task /path/to/img1.jpg /path/to/img2.jpg ... /path/to/imgN.jpg > result.json

HTTP API

HTTP API is based on Falcon, so it can be used with any WSGI server, such as uWSGI or Gunicorn.

• install your favourite WSGI server (e.g. pip install gunicorn)
• set env variable CEEVEE_TASKS for your tasks, multiple comma separated tasks are supported, e.g. CEEVEE_TASKS=task1,task2
• run a server CEEVEE_TASKS=dummy gunicorn ceevee.cv_http;
• send a POST request with image parameter.

$ http -f POST localhost:8000/dummy image@/tmp/img.jpg
HTTP/1.1 200 OK
Connection: close
Date: Sat, 14 Sep 2019 13:47:39 GMT
Server: gunicorn/19.9.0
content-length: 37
content-type: application/json

{
    "result": [
        500,
        500,
        3
    ],
    "success": true
}

Python API

from ceevee.utils import read_img
from ceevee.dummy import DummyPredictor
baseline = DummyPredictor()
img = read_img('/path/to/img.jpg')
result = baseline(img)

Contributions

Yes, you can add a new model!

Checklist:

• create a GitHub issue with your suggested model;
• create a new Baseline class (see ceevee.dummy.DummyBaseline) and implement three methods (preprocess, process, postprocess);
• add your model to MODELS at ceevee/__init__.py
• add tests to tests/;
• once CI is green, create a pull request!

ToDo:

• infrastructure:
  • packaging, pip
• APIs:
  • http: tests, error handling
• models:
  • face detection
  • face emotion
  • face keypoints
  • car detection
  • crowd density estimation