akella17 Goto Github PK

akella17.github.io

beta-vae

To learn and reason like humans, AI must first learn to factorise interpretable representations of independent data generative factors (preferably in an unsupervised manner!!). What does all this mean? Go through this tutorial to get an overview of disentanglement in the context of unsupervised visual disentangled representation learning.

deep-bayesian-quadrature-policy-optimization

Official implementation of the AAAI 2021 paper Deep Bayesian Quadrature Policy Optimization.

enhancenet

Achieves realistic textures by using automated texture synthesis in combination with a perceptual loss rather than focusing on optimizing for a pixel accurate reproduction of ground truth images during training. By using feed-forward fully convolutional neural networks in an adversarial training setting, this approach achieves a significant boost in image quality at high magnification ratios.

handwriting_synthesis

This work attempts to generate sequences of handwritten sentences using LSTM network and Mixture Model (Based on the work : https://arxiv.org/pdf/1308.0850.pdf by Alex Graves)

language_identification

This is an implementation of a character-level LSTM network for language identification. Inspired from Stanford Language Identification Engine(SLIDE) : https://arxiv.org/abs/1701.03682

open3d

Open3D: A Modern Library for 3D Data Processing

seqqlearning

speaker-embedding

A deep neural network for finding text-independent speaker embedding written in tensorflow and tensorpack

speaker-recognition

TensorFlow implementation of a 3 Layer Stacked LSTM architecture to classify speakers in VCC (2016) dataset.

voice_style_transfer

Attempts to perform voice transfer, inspired by Gatys et al.'s work in image domain. Uses two pre-trained networks (Wavenet and Speaker Recognition) for perceptual style and context losses.