We present an analysis of embeddings extracted from different pre-trained models for content-based image retrieval. Specifically, we study embeddings from image classification and object detection models. We discover that even with additional human annotations such as bounding boxes an