Evolution of the Diffusion model
Diffusion hasn’t always been available, just as Rome was not built in a day. To have a high-level bird’s view of this technology, in this section, we will discuss the overall evolution of the Diffusion model in recent years.
Before Transformer and Attention
Not too long ago, Convolutional Neural Networks (CNNs) and Residual Neural Networks (ResNets) dominated the field of computer vision in machine learning.
CNNs and ResNets have proven to be highly effective in tasks such as guided object detection and face recognition. These models have been widely adopted across various industries, including self-driving cars and AI-driven agriculture.
However, there is a significant drawback to CNNs and ResNets: they can only recognize objects that are part of their training set. To detect a completely new object, a new category label must be added to the training dataset, followed by retraining or fine-tuning the pre-trained models.
...
