meme-understanding

Raw datasets

Link for the raw datasets:

• MAMI: https://competitions.codalab.org/competitions/34175#learn_the_details
• Hateful memes: https://ai.facebook.com/blog/hateful-memes-challenge-and-data-set/

Setting up environment

• M1
• x86 Linux
• Colab

To extract data

Run following scripts from repo root directory to extract data into ./data/extracted for MAMI and ./data/extracted_hateful_meme for hateful meme dataset

• sh scripts/extract_mami_data.sh <raw_data_path>
• sh scripts/extract_hateful_meme.sh <path/hateful_memes.zip>

This repo supports

• Meme classification using neural network src/runner/classify_mami.py
• Meme classification using xDNN src/runner/classify_xdnn.py
• Meme classification and similar feature extraction src/runner/mami_knn_clip_runner.py

Experiments

The experiments were ran through vscode launch config, please refer the same as the execution instructions.