augment.py

import os
import numpy as np
from scipy.ndimage import shift, rotate

DATA_PATH = os.path.join('frontend', 'models')

def augment_data():
    print(f"⚡ Generating Clones in {DATA_PATH}...")
    files = [f for f in os.listdir(DATA_PATH) if f.endswith('.npy') and "aug" not in f]
    
    count = 0
    for file_name in files:
        file_path = os.path.join(DATA_PATH, file_name)
        try:
            data = np.load(file_path) # Shape: (40, 225)
            
            # 1. NOISE CLONE (Simulates bad webcam grain)
            noise = np.random.normal(0, 0.01, data.shape)
            data_noise = data + noise
            np.save(os.path.join(DATA_PATH, file_name.replace('.npy', '_aug_noise.npy')), data_noise)
            
            # 2. SHIFT CLONE (Simulates person standing slightly left/right)
            # Shift x-coordinates (indices 0, 3, 6...)
            data_shift = data.copy()
            data_shift[:, 0::3] += 0.02 # Shift Right
            np.save(os.path.join(DATA_PATH, file_name.replace('.npy', '_aug_shift.npy')), data_shift)
            
            # 3. SPEED CLONE (Simulates faster signing)
            # We skip every 2nd frame to make it "faster" then pad with zeros
            data_fast = np.zeros_like(data)
            fast_seq = data[::2] # Take every other frame
            data_fast[:len(fast_seq)] = fast_seq
            np.save(os.path.join(DATA_PATH, file_name.replace('.npy', '_aug_fast.npy')), data_fast)

            count += 3
        except Exception as e:
            print(f"Skipped {file_name}: {e}")

    print(f"✅ Created {count} new training samples!")

if __name__ == "__main__":
    augment_data()