The MUSTI challenge @ MediaEval 2023 - Multimodal Understanding of Smells in Texts and Images with Zero-shot Evaluation

We ran the MUSTI challenge the second time after the MUSTI 2022 edition by extending the evaluation with a zero-shot evaluation scenario. This was needed as the first iteration showed us there is a lot of room for improvement and zero-shot performance of the state-of-the-art methods is useful in understanding what available models can predict without any training in a new language. We used the same data from MUSTI 2022 for training and evaluation for MUSTI 2023. Additionally, we prepared a second evaluation scenario, which we call zero-shot, in Slovenian, which was not known by the participants before the evaluation phase started. MUSTI 2023 has attracted many teams and state-of-the-art multimodal systems perform better than the systems proposed in MUSTI 2022.