What is Embedding and How Does it Work?
How Image Comparison Using Embeddings Works
00:00 Introduction
00:32 What is embedding and how does it work?
03:26 When to use embedding
05:37 Examples of use
07:38 Dataset organization
10:27 How training works
13:11 Loss functions
17:11 Comparison and threshold calculations
19:51 Conclusion
↔️ Embedding transforms each image into a numerical vector, and by comparing these vectors, we can measure how similar two images are. This allows us to decide if we are looking at the same object or not! They enable us to perform comparisons, whether images or words. They can be used in various applications, such as facial recognition, searches for similar images, and even product recommendations, facilitating the process of finding matches and patterns.
🤖 In this video, we will explore embedding models and how they can be used to compare two images. Whether it's to identify if they belong to the same person, object, or for any other need, embeddings are a powerful tool. We'll see how they work, what the main differences are compared to classifiers, and in which situations they are most suitable.
✨ We will go through details and practical examples of use, as well as one of the possibilities for organizing the data in a dataset, how training works, some commonly used loss functions, and the preparations made to perform inferences. It's not enough to train these models; we need to take the generated vectors, perform the calculations to convert them into a value that we can use, through a threshold, to identify if the images contain the same object.
🌐 References:
▶ DeepFace Repository - https://github.com/serengil/deepface
▶ More information on embedding and calculations - / face-embedding-and-what-you-need-to-know
#Embeddings #EmbeddingModels #ArtificialProgrammer