trellixvulnteam/aspectus_pkwr

Implementation of Image Segmentation and Classification using Python on Tensorflow v0.12 and wrapper Slim, neural net VGG-16s, Scikit Image Library and Inception. The updated version of Aspectus contains custom made sticker batches for Telegram using the output.png file

Aspectus

Copyright (C) 2017 Aakash Sinha [email protected]

Credits : @warmspringwinds

We will be performing image segmentaion on a given image. The machine learning library which we are using is TensorFlow. In the first phase we used Slim wrapper and VGG‐16 model to classify an image in over 1000 classes and give significant probabilities using Softmax. In the second phase we have trained FCN‐8s net on VGG‐16 and it has used PASCAL VOC 2012 model (trained on ImageNet to generate 21 classes) to generate classes. Then we have used CRF as recursive function (in RNN) to generate a heat map of the obtained foreground whic have been classified. Then we have used morphological operations on the image to detect contour and masking to retrieve the final output image. Then we have extended our project by making custom stickers for Telegram ChatBox.