Módszerek összehasonlítása

	Multilingvis képosztályozás	Multimodális képbesorolás
Tudományterület	Mélytanulás	Mélytanulás
Módszercsalád	Machine learning	Machine learning
Keletkezés éve	2020s	2011–2021
Megalkotó	Community / Radford et al. (CLIP, 2021) as key enabler	Ngiam et al.; Radford et al. (CLIP)
Típus	Cross-lingual supervised image classification	Multimodal supervised classification
Alapmű	Radford, A., Kim, J. W., Hallacy, C., Ramesh, A., Goh, G., Agarwal, S., ... & Sutskever, I. (2021). Learning transferable visual models from natural language supervision. In Proceedings of the 38th International Conference on Machine Learning (ICML), pp. 8748–8763. PMLR.	Radford, A., Kim, J. W., Hallacy, C., Ramesh, A., Goh, G., Agarwal, S., ... & Sutskever, I. (2021). Learning transferable visual models from natural language supervision. Proceedings of the 38th International Conference on Machine Learning (ICML), PMLR 139, 8748–8763.
Alternatív nevek	Cross-lingual image classification, Multilingual visual recognition, Cross-cultural image classification, Multilingual vision-language classification	multimodal visual classification, image-text classification, vision-language classification, cross-modal image classification
Kapcsolódó	5	6
Összefoglaló	Multilingual image classification trains visual models to recognise and label images when class names, supervision signals, or evaluation benchmarks span multiple languages. Enabled by multilingual vision-language models such as CLIP, it allows a single model to classify images using prompts or labels in any supported language, facilitating cross-cultural and cross-lingual deployment of computer vision systems.	Multimodal image classification extends standard visual classification by incorporating additional modalities — such as text captions, audio, or structured metadata — alongside image features. Separate encoders process each modality, their representations are fused, and a joint classifier assigns the target label. Models such as CLIP demonstrate that image–text alignment enables zero-shot and few-shot image classification at scale.