https://github.com/activestate/tensorflask
Simple example hosting TensorFlow model as Flask service
https://github.com/activestate/tensorflask
Last synced: 9 months ago
JSON representation
Simple example hosting TensorFlow model as Flask service
- Host: GitHub
- URL: https://github.com/activestate/tensorflask
- Owner: ActiveState
- License: apache-2.0
- Created: 2017-11-27T17:47:22.000Z (over 8 years ago)
- Default Branch: master
- Last Pushed: 2023-02-08T04:41:20.000Z (over 3 years ago)
- Last Synced: 2025-04-05T04:51:15.385Z (about 1 year ago)
- Language: Python
- Size: 4.72 MB
- Stars: 20
- Watchers: 24
- Forks: 14
- Open Issues: 24
-
Metadata Files:
- Readme: README.md
- License: LICENSE
Awesome Lists containing this project
README
# tensorflask
This is a simple example of hosting a TensorFlow model as Flask service for inference. It provides the "Poodle, Pug or Weiner Dog?" image identification service using a retrained MobileNet model. The retrained model/labels are provided here to let you run the service locally.
## Setup
1. You need Python 3.x with Flask and TensorFlow installed. You can [download ActivePython 3.5](https://www.activestate.com/activepython/downloads) which has all the required dependencies already pre-installed.
2. Clone to repository by clicking the clone button above.
3. Run `python app.py`.
## Usage
Once you've started the service, you can query it on `localhost:8000`. You can either hit it via a web browser, or use `curl` from the commandline. It takes a single parameter `file` which specifies the full path to a local image, so for example:
`curl http://localhost:8000?file=/home/pete/mypoodle.jpg`
Will send the photo to the service, which will run the model and return JSON identifying the probabilties of each type of dog breed. In this case you'll get results like:
```json
[
[
"poodle",
"pug",
"dachshund"
],
[
0.9994891881942749,
1.1696176443365403e-05,
0.0004991634050384164
]
]
```
And you can see that the model is 99% sure that the image is a poodle.
## License
Licensed under the Apache 2.0 license. See LICENSE file for details.