Blogs

Resnet Strikes back

Resnet Strikes Back

Understand how the model, training architecture and randomness are stitched together in deep learning. This blog contains notes from the amazing paper Resnet strikes back by Wightman et al.

keep reading ...
Vishnu Subramanian
...

How to train models on multiple GPUs using fastai

Quick start guide to train deep learning models in multi-GPU setup using fastai

keep reading ...
Vishnu Subramanian
...

Understanding and Building Resnets from scratch using Pytorch

Resnets are the go-to architecture for most of the deep learning tasks. What makes them so popular? Resnets introduced residual blocks which made deeper neural networks possible. In this post, Poonam Ligade shares understanding of resnets and shows how you can build Resnet from scratch using Pytorch.

keep reading ...
Poonam Ligade
...

Identifying people without Masks - Using Deep Learning/AI

Vishnu Subramanian takes you through high level steps required for building an object detection pipeline for identifying persons without masks. He also discusses the challenges faced with this approach. Along with that he shares how we can use deep learning algorithms to help us in semi-automating the labelling task.

keep reading ...
Vishnu Subramanian
...

Building image segmentation algorithm using fastai2/PyTorch
Part - 1

The first post in series introduces the problem of identifying salt deposits beneath the earth, describes why we choose fastai2 and build a data pipeline required for training a segmentation algorithm.

keep reading ...
Vishnu Subramanian
...

Building image segmentation algorithm using fastai2/PyTorch
Part - 2

Learn about how to build a unet model for image segmentation with a custom encoder. Along the way you will build a solution which ranks in the top 4% of kaggle leaderboards. We will also create a pipeline required for predicting the test data, applying custom TTA and generating a kaggle submission file.

keep reading ...
Vishnu Subramanian
...

Understanding Autoencoders and Variational Autoencoders

In the recent years, computer vision algorithms have achieved many things. One amazing and dangerous thing they can do is generate new images, faces, voices, etc. Let's look at the fundamental techniques that power modern generative, segmentation models and much more.

keep reading ...
Vishnu Subramanian
...

7 Tips and tricks for transfer learning in Pytorch

Transfer learning has become a key component of modern deep learning, both in the fields of CV and NLP. Read the post to learn how to tweak your neural network to achieve better results from your data.

keep reading ...
Vishnu Subramanian
...

From buying to building AI workstations

Read Vishnu Subramanian talk about his deep learning journey along with some of the key challenges faced by companies while prototyping machine learning models. Challenges range from picking right hardware to installing and managing vibrant software stack. Learn how Jarvislabs.ai is solving those problems with Jarvis Manager and Jarvis Workstations.

keep reading ...
Vishnu Subramanian