Karthik Tech Blogs

EfficientDet

Paper: EfficientDet This paper discusses about the optimized approach in Image detection tasks and compares it with other detection models such as YOLOv3, MaskRCNN and many more by building a scalable detection architecture. It proposes a weighted Bi-directional feature pyramid network (BiFPN), which allows easy and fast multi-scale feature ... Read more

Decoding gradients

This blog provides indepth understanding of optimization from Introduction to deep learning course by CMU . Course Page In neural networks, the convergence is attained by finding the optimal network error. The error is defined as a function called the Loss function, which must be minimized through an iterative process of weight updates... Read more

GANS

Generative Adversarial Nets A generative model G that captures the data distribution. A discriminative model D estimates the probability that a sample came from the training data rather than G. Dis_probability = ( Number of samples from training data)/( Total Number of Samples generated by generator) The training procedure for Generator is ... Read more

4D Spatio-Temporal Convnets

Paper: 4D Spatio-Temporal ConvNets In many robotics and VR/AR applications, 3D-videos are readily-available sources of input (a continuous sequence of depth images, or LIDAR scans). However, those 3D-videos are processed frame-by-frame either through 2D convnets or 3D perception algorithms. In this work, we propose 4-dimensional convolution... Read more

U-GAT-IT Architecture

Paper: U-GAT-IT: Unsupervised Generative Attentional Networks with Adaptive Layer-Instance Normalization for Image-to-Image Translation Architecture: Generator: Let x ∈ {Xs, Xt} represent a sample from the source and the target domain. Our translation model Gs→t consists of an encoder Es, a decoder Gt, and an auxiliary class... Read more

Digital Matting

Paper: A Late Fusion CNN for Digital Matting This paper studies the structure of a deep convolutional neural network to predict the foreground alpha matte by taking a single RGB image as input. Our network is fully convolutional with two decoder branches for the foreground and background classification respectively. Then a fusion branch is ... Read more

Class Activation Map

Learning Deep Features for Discriminative Localization We know that Convolution Neural Networks are good at classification tasks. This paper decodes how previous layer activation contribute for localization tasks even though the network is being trained on classification tasks. Using Global Average pooling layer, the localization ability... Read more

AI @ fb

From Satellite Imagery to Disaster Insights Large-Scale Visual Relationship Understanding Spatially Invariant Unsupervised Object Detection with Convolutional Neural Networks SGD Implicitly Regularizes Generalization Error Rethinking floating point for deep learning A2 -Nets: Double Attention Networks Explore-Exploit:... Read more

All about Seq2seq

Sequence to Sequence Learning Sequence to sequence model comprises of two sequence networks one at the encoder side and other at the decoder side. These sequence networks are RNN. The encoder networks maps the input sequence to a vector of a fixed dimension. The decoder network decodes the target sequence from the encoded vector. Seq2seq... Read more