RandomOutline

Data Augmentation with Random Outline Processing

About

Fill outline with random color with parameter p. Bigger p, Larger outline size.

Trained models are here.

All jupyter notebook codes are here

import

import numpy as np
import time
import random
import torch
import torch.nn as nn
import torch.nn.functional as F
from torch.autograd import Variable
import torch.backends.cudnn as cudnn
from torch.optim.lr_scheduler import MultiStepLR

from torchvision import datasets, transforms

from tqdm.notebook import tqdm as tqdm

transforms

train_transform = transforms.Compose([])
train_transform.transforms.append(transforms.RandomCrop(32, padding=4))
train_transform.transforms.append(transforms.RandomHorizontalFlip())
train_transform.transforms.append(transforms.ToTensor())
train_transform.transforms.append(normalize)
train_transform.transforms.append(RandomOutline(0.05))

result

Best Top-1 Accuracy: 77.84

Best Top-5 Accuracy: 94.13

result outputs are here.