CNN Image Retrieval Learns from BoW: Unsupervised Fine-Tuning with Hard Examples
This addresses the need for automated fine-tuning in computer vision, particularly for object retrieval, but it is incremental as it builds on existing retrieval and SfM methods.
The paper tackles the problem of reducing manual annotation for CNN fine-tuning in image retrieval by proposing an unsupervised method that uses 3D models from retrieval and SfM to select training data, resulting in enhanced performance with compact codes.
Convolutional Neural Networks (CNNs) achieve state-of-the-art performance in many computer vision tasks. However, this achievement is preceded by extreme manual annotation in order to perform either training from scratch or fine-tuning for the target task. In this work, we propose to fine-tune CNN for image retrieval from a large collection of unordered images in a fully automated manner. We employ state-of-the-art retrieval and Structure-from-Motion (SfM) methods to obtain 3D models, which are used to guide the selection of the training data for CNN fine-tuning. We show that both hard positive and hard negative examples enhance the final performance in particular object retrieval with compact codes.