(NOTE: Run times may vary depending on your Internet connection and computer's CPU speed.)

Once TensorPy is installed with install.sh, a command called classify is added to your command line, which takes a URL as input. When called, TensorPy determines if that URL links to an image or a web image. If it's an image, TensorPy downloads the image to a new folder called downloads_folder. From there, the image is converted to a JPEG. Then, the image is fed to the local tensorpy/classify_image.py which tells TensorFlow to use the ImageNet Inception database to classify the image and print out the result.

If the classify command is called with a web page as input, then all images on that page (up to the limit defined in tensorpy/settings.py) will get downloaded to downloads_folder, where TensorPy does the work listed above per image, with the exception of images smaller than the minimum size defined in tensorpy/settings.py (currently 50x50 pixels). Images that are too small will be skipped because results have shown that icons and other tiny images get extremely poor classification results.

You can also use TensorPy to classify images from within a Python program. (See the example to learn how: examples/test_python_classify.py)

If you make any changes to tensorpy/settings.py, you will need to rerun python setup.py install for your changes to take effect.