MEMES4GOOD-MODELS

Memes4Good Models

This repository contains the three versions of the Memes4Good meme classifier:

• ROBUST
• BALANCED
• POWERFUL

These models were trained on Spanish memes using a multimodal architecture combining image and text.

Feel free to reach out if you have any questions or suggestions!<br> <br> Links:

• LinkedIn

• GitHub - Project & Notebooks – here you can find the notebooks and the full README explaining the project in detail.

• Hugging Face - Dataset – all the memes used for training and validation.

• GitHub - Metadata CSV – CSV containing key columns used for training and inference:

  • meme_id: Unique identifier corresponding to the meme image filename.
  • extracted_text: The literal text present in the meme image.
  • description_es_new: A Spanish semantic description of the visual scene and context.
  • harmless: The target label where 0 indicates harmful/offensive content and 1 indicates harmless content.