Unsupervised Domain Adaptation for Image Classification Using Non-Euclidean Triplet Loss

In recent years, computer vision tasks have increasingly used deep learning techniques. some tasks, however, due to insufficient data, the model is not properly trained, leading a decrease in generalizability. When trained on dataset and tested another similar dataset, predicts near-random results. This paper presents an unsupervised multi-source domain adaptation that improves transfer increases proposed method, new module infers source of input data based its extracted features. By making features extractor compete against this objective, learned feature representation generalizes better across sources. As result, representations those from different sources are learned. That is, generic independent any particular domain. training stage, non-Euclidean triplet loss function also utilized. Similar for samples belonging same class can be more effectively using function. We demonstrate how developed framework may applied enhance accuracy outperform outcomes already effective methodologies. strategy performs particularly well when dealing with various domains or there data.