SAFE-MEME is a novel framework designed for the detection of fine-grained hate speech in memes. The framework consists of two distinct variants: (a) SAFE-MEME-QA, which employs a Q&A approach, and (b) SAFE-MEME-H, which utilizes hierarchical classification to categorize memes into one of three classes: explicit, implicit, or benign.
General Instruction:
-
Create a conda environment: conda create --name env_X
-
Activate the environment: conda activate env_X
-
Please install amazon-science/mm-cot: https://github.com/amazon-science/mm-cot
-
Run pip install -r requirements.txt
-
Chnage directory to mm-cot
-
Add or replace the following file (folder) in mm-cot folder, ** timm, vision_features, unit_models (can be trained from scratch too) ** use: https://drive.google.com/drive/folders/1PWESNhZIDa1YL6aipVyrl_Pwo4LYo1Zu?usp=sharing
-
Please put the all the .py files from pyFiles folder in mm-cot folder.
-
Create a folder named, 'results' and 'resultsConf'