cvaisnor / minecraft-biome-detection Goto Github PK

• OS: Ubuntu 22.04.2 LTS
• Python Version: 3.10.12
• PyTorch Version: 2.1.0+cu121

• biome_videos/: Contains the raw videos for the dataset

Original Video Structure
- /biome_videos
    - /category1
        - video.mkv
    - /cateogry2
        - video.mkv
    - ....

• This format should work if multiple videos are present for each category.
• preprocessing.py: Script to convert videos to frames and store them in the following structure

Frame Storage Structure
- /frames
    - train
        - /category1
            - frame0.jpg
            ....
            - frameN.jpg
        - /category2
        - ....
    - test
        - /category1
            - frame0.jpg
            ....
            - frameN.jpg
        - /category2
        - ....

Example Usage:

python preprocessing.py -dd biome_videos -fd frames -nf 200

Frames are uniformly sampled from each video to create the dataset. The dataset is split into 80% training and 20% testing. Once the frames are stored in the correct structure, you can use main.ipynb to walk through the process of training and testing the model.

• Various ResNet models can be initialized with pretrained weights from the ImageNet dataset.

Transfer learning and fine-tuning is used to train the model on 10 distinct Minecraft biomes. The dataset consists of 10 classes:

• Desert
• Birch Forest
• Jungle
• Crimson Forest
• Dark Forest
• End Midlands
• Savanna
• Snowy Taiga
• Swamp
• Plains.