Advanced Data Augmentation

Module: Computer Vision | Difficulty: Intermediate

Mixup

where .

CutMix

where is a binary mask and .

RandAugment

Apply random operations from a pool with magnitude :

AutoAugment

Learn augmentation policy via reinforcement learning.

import torch
import numpy as np

def mixup_data(x, y, alpha=0.2):
    lam = np.random.beta(alpha, alpha)
    batch_size = x.size(0)
    index = torch.randperm(batch_size)
    mixed_x = lam * x + (1 - lam) * x[index]
    return mixed_x, y, y[index], lam

def cutmix_data(x, y, alpha=1.0):
    lam = np.random.beta(alpha, alpha)
    batch_size = x.size(0)
    index = torch.randperm(batch_size)
    
    _, _, H, W = x.shape
    cut_ratio = np.sqrt(1 - lam)
    rw = int(W * cut_ratio)
    rh = int(H * cut_ratio)
    cx = np.random.randint(W)
    cy = np.random.randint(H)
    
    x1, y1 = max(0, cx - rw // 2), max(0, cy - rh // 2)
    x2, y2 = min(W, cx + rw // 2), min(H, cy + rh // 2)
    
    mask = torch.ones_like(x)
    mask[:, :, y1:y2, x1:x2] = 0
    
    mixed_x = mask * x + (1 - mask) * x[index]
    lam = 1 - (x2 - x1) * (y2 - y1) / (W * H)
    return mixed_x, y, y[index], lam

Key Takeaways

• Mixup and CutMix improve calibration and generalization
• RandAugment is simpler and competitive with AutoAugment
• Augmentation should match the domain and task