PyTorch

R-CNN Traditional Convolutional Neural Networks (CNNs) with fully connected layers often struggle with object detection tasks, especially when dealing with multiple objects of various sizes and positions within an image. A brute-force method like applying a Sliding Window (Exhaustive Search) across the image to detect objects is highly computationally expensive, as it fails to scale efficiently when object frequency and variation increase. Regions with CNN features (R-CNN) [1] was introduced in 2014 to overcome these challenges. R-CNN presents an approach by using a Selective Search algorithm to generate around 2,000 region proposals from an image. These proposals are likely to contain objects and are individually processed to detect and localize objects more efficiently. R-CNN marked a significant advancement in the field of object detection and laid the foundation for faster and more accurate object detection models. ...

Convolutional neuron networks (CNN) have been revolutionized image-related Artificial Intellegence (AI) tasks. This kind of neuron network architectures have been obtained huge success. Furthermore, CNNs are very interesting. They have many design elements and variants you could play with. Academic researches might pursue a beauty of purity and emphasize the unique feature they proposed. However, in industry application, it’s not necessary to build a network pure. It’s highly possible to combine all necessary design elements together. However, we should have a deep understand of all these design elements. ...

While training a model, a few important mathematical tricks have to be employed. Normalization and regularization techiniques belong to them. They are very important to successfully train models, and they are often confusing and sometimes hard to understand the details as well. Additionally, learning rate scheduling is also very important while training models, especially large ones. Normalization vs Regularization To put it simple, normalization is applied on data. Before training a model, data should be normalized. There are a few irresistible benefits to do it. It stablizes the training process, speeds up model convergence significantly, and improve model’s generalization capability. ...