Opinion and sentiment analysis is a vital task to characterize subjective\ninformation in social media posts. In this paper, we present a comprehensive\nexperimental evaluation and comparison with six state-of-the-art methods, from\nwhich we have re-implemented one of them. In addition, we investigate different\ntextual and visual feature embeddings that cover different aspects of the\ncontent, as well as the recently introduced multimodal CLIP embeddings.\nExperimental results are presented for two different publicly available\nbenchmark datasets of tweets and corresponding images. In contrast to the\nevaluation methodology of previous work, we introduce a reproducible and fair\nevaluation scheme to make results comparable. Finally, we conduct an error\nanalysis to outline the limitations of the methods and possibilities for the\nfuture work.\n