Generative models for open set image recognition

• Autor(a) principal: Marcos Felipe Vendramini Carvalho
• Data de Publicação: 2021
• Tipo de documento: Dissertação
• Idioma: por
• Título da fonte: Repositório Institucional da UFMG
• Texto Completo: http://hdl.handle.net/1843/49031
• topic: Computer vision; Neural networks; Open set recognition; Computação – Teses; Visão computacional –Teses; Redes neurais (Computação) – Teses; Reconhecimento de conjunto aberto –Teses.

Image classification methods are usually trained to perform predictions taking into account a predefined group of known classes. Real-world problems, however, may not allow for a full knowledge of the input and label spaces, making failures in recognition a hazard to deep visual learning. Open set recognition methods are characterized by the ability to correctly identify inputs of known and unknown classes. In this context, we propose GeMOS: simple and plug-and-play open set recognition modules that can be attached to pre-trained Deep Neural Networks for visual recognition. The GeMOS framework pairs pre-trained Convolutional Neural Networks with generative models for open set recognition to extract open set scores for each sample, allowing for failure recognition in object recognition tasks. We conduct a thorough evaluation of the proposed method against state-of-the-art open set algorithms. In these tests, different datasets were used, such as in and out of distribution, with the MNIST as in distribution, we reached the F1-score of 0.91 while the best baseline of the referent test reached 0.85, and, for the CIFAR10 as in distribution, we reached the F1-score of 0.93 while the best baseline of the benchmark test reached 0.81. Tests were also performed using the same dataset as in and out of distribution, a more complex case that showed the dependence of the method on the accuracy of pre-trained networks. The results showed that GeMOS competes with more complex and expensive models and in many cases outperforms them. For future work, we initially propose to apply the method to other domains and real-world problems, and to modify the method for other computer vision tasks.