Neural network embeddings are a low-dimensional representation of input data that gives rise to various applications. Embeddings have some interesting capabilities, as they can capture the semantics of the data points. This is especially useful for unstructured data like images and videos so that you can encode not only pixel similarities but also some more complex relationships.
Embeddings from the BDD100K dataset were visualized using FiftyOne and Plotly.
Performing searches over these embeddings gives rise to many use cases like classification, building up recommendation systems, or even anomaly detection. One of the primary benefits of performing a nearest-neighbor search on embeddings to accomplish these tasks is that there is no need to create a custom network for every new problem; you can often use pre-trained models. Using the embeddings generated by some publicly available models is possible
without any further finetuning.