tf_img2vec

This library uses the ResNet50 model in TensorFlow Keras, pre-trained on Imagenet, to generate a 2048 embedding vector for an input image.

Install

1. With a local version of the code:

git clone https://github.com/Lambert-Shirzad/tf_img2vec

cd ./tf_img2vec

pip install --user ./

2. Directly from github:

pip install git+git://github.com/Lambert-Shirzad/tf_img2vec.git

Usage

from tf_img2vec.img2vec import Img2Vec
img2vec = Img2Vec()
x = img2vec.get_vec('./examples/tree1.jpeg')

Reference

Majority of the code is taken from https://github.com/jaredwinick/img2vec-keras The main updates are:

1. Initialization of the embedding model:

I have used the built-in Keras functions to directly load a ResNet50 model ending at the avg_pool layer. This make the model slightly lighter and faster to load.

2. Downloading of the pre-trained model:

The script was changed to download the weights for the clipped version of ResNet50 into a local ./model directory.