Human_Activity_Video_Recognition

• Add hugging face models
• Add proper model logging
• Train model properly
• Add elas names to confusion matrix
• Add training history plotting
• Add training history comparison dataset
• No EARLY STOP HMDB
• No EARLY STOP UFC
• No EARLY STOP CUSTOM DATASET
• Look into generator for input data, to lower memory issues
• Look into GPU training setup
• Implement single class prediction
• Implement multi class multi class confidence percentage prediction
• Look into model architecture changes
• Look into regularizers L1, L2, Dropout and early stopping,
• Implement camera feed recognition
• Add grid search training
• Implement CLI

transfer learning

Usage

Help

python main.py --help

Training

python main.py --functionality train --model convlstm --dataset dvorak_custom --epochs 50 --test-percentage 0.2 --resolution 224 --batch-size 32

python main.py --functionality train --model resnet --dataset dvorak_custom --epochs 50 --test-percentage 0.2 --resolution 224 --batch-size 32

python main.py --functionality train --model googlenet --dataset dvorak_custom --epochs 50 --test-percentage 0.2 --resolution 224 --batch-size 32

python main.py --functionality train --model timesformer --dataset dvorak_custom --epochs 50 --test-percentage 0.2 --resolution 224 --batch-size 32

Prediction

python main.py --functionality predict --single-class-prediction --model MODEL_TYPE --model-path "MODEL_PATH.keras" --paths "['VIDEO_PATH.mp4']"