Walking on the intersection of research and programming
All of these generative models ultimately derive from Maximum Likelihood, at least implicitly. The variational autoencoder sits in the “explicit” part of the tree. Remember that we had a clear loss function (the reconstruction loss)? Well, with GANs we do not have it anymore. Rather, we now have two competing loss functions that we will … Continue reading Training & Common Challenges: GANing for Success
Why Semi-Supervised Learning? Semi-supervised learning in one of the most promising areas of practical application of GANs. Unlike supervised learning, where we need a label for every example in our dataset, and unsupervised learning, where no labels are used semi-supervised learning has a class for only a small subset of example. The lack of labeled … Continue reading Semi-Supervised GAN
In this chapter, we want to provide you with hands-on tutorial to build a Progressive GAN (aka PGGAN or ProGAN) using TensorFlow and the newly released TensorFlow Hub (TFHub). The progressive GAN is a cutting-edge technique that was published at ICLR 2018 and has manage to generate full-HD photo-realistic images, or smoothly combine any of … Continue reading Progressing with GANs
Among the many applications of GAN, image synthesis is the most well-studied one, and research in this area has already demonstrated the great potential of using GAN in image synthesis. In this paper, the author provide a taxonomy of methods used in image synthesis, review different models for text-to-image synthesis and image-to-image translation, and discuss … Continue reading Paper Daily: An Introduction to Image Synthesis with Generative Adversarial Nets
With the availability of low-cost and compact 2.5/3D visual sensing devices, computer vision community is experiencing a growing interest in visual scene understanding of indoor environments. This survey paper provides a comprehensive background to this research topic. We begin with a historical perspective, followed by popular 3D data representations and a comparative analysis of available … Continue reading Paper Daily: Indoor Scene Understanding in 2.5/3D for Autonomous Agents: A Survey
In this paper, the authors examine a collections of training procedure and model architecture refinements that improve model accuracy but barely change computational complexity. Many of them are minor “tricks” like modifying the stride size of a particular convolution layer or adjusting learning rate schedule. Collectively, however, they make a big difference. Training Procedures Baseline … Continue reading Paper Daily: Bag of Tricks for Image Classification with Convolutional Neural Networks
The most challenging algorithmic problems involve optimization, where we seek to find a solution that maximizes or minimizes some function. Algorithms for optimization problems require proof that they always return the best possible solution. Greedy algorithms that make the best local decision at each step are typically efficient by usually do not guarantee global optimality. … Continue reading Warm-up Coding Interview: Dynamic Programming
In this section, we introduce backtracking as a technique for listing all possible solutions for a combinatorial algorithm problem. We illustrate the power of clever pruning techniques to speed up real search applications. For problems that are too large to contemplate using brute-force combinatorial search, we introduce heuristic methods such as simulated annealing. Such heuristic … Continue reading Warm-up Coding Interview: Combinatorial Search and Heuristic Methods
There is an alternate universe of problems for weighted graphs. The edges of road networks are naturally bound to numerical values such as construction cost, traversal time, length, or speed limit. Identifying the shortest path in such graphs proves more complicated than breadth-first search in unweighted graphs, but opens the door to a wide range of … Continue reading Warm-up Coding Interview: Weighted Graph Algorithms
In this paper, the authors proposed a long-term feature bank – supportive information extracted over the entire span of a video – to augment state-of-the-art video models that otherwise would only view short clips of 2-5 seconds.