Should you annotate your data? As an engineer whose work concerns learning-based architecture in the field of robotics and automation, data generation is one of the easier challenges to overcome. But it is one thing to generate data in heaps and piles and a completely different thing to get... [Read More]

We looked at the process of image segmentation and its different types in my previous blog. In this post, we are going to take a deeper look at the architecture of U-Net. We will also discuss the PyTorch-based code for U-Net. The complete code can be found <a href="https://github.com/nikheelpandey/collections/tree/main/SemanticSeg/unet/" target="_blank"... [Read More]

The object detection and classification task make the most buzz in the field of computer vision. But one of the most important tasks in computer vision is segmentation. In this blog, I’ll be detailing the application of segmentation-related tasks and in my next blog, I’ll introduce you to an instrumental... [Read More]

Usually, when we train a model on a classification task, we are making the model learn the distinction between classes inadvertently. However, we end up optimizing for learning global features that are useful for a high-level understanding. These models would typically fail to generalize for tasks for which they are... [Read More]

One of the best papers I came across in 2020 was the Google Brain team’s Self-training with Noisy Student. It is a semi-supervised technique that works well when we have an insufficient amount of labeled data. So much so that it beat the contemporary... [Read More]