Awesome
ImageRecognitionDataset
Caltech101/256, CIFAR-10/100, MNIST/FashionMNIST, omniglot
Requirement
- Python >= 3.8
- Poetry >= 1.2
Install
pip
pip install numpy pillow tqdm
poetry
poetry install
Usage
# Dataset Download
python src/download.py --dataset {CIFAR10 | CIFAR100 | MNIST | fashionMNIST | caltech101 | caltech256 | omniglot}
# Calculate Dataset Mean Std
python src/calculate.py --dataset {CIFAR10 | CIFAR100 | MNIST | fashionMNIST | caltech101 | caltech256 | omniglot}
Caluculated Result
GrayScale dataset
| dataset | mean | std |
|---|---|---|
| MNIST(train) | 0.1307 | 0.3013 |
| fashionMNIST(train) | 0.2860 | 0.3202 |
| Omniglot(images_background) | 0.9221 | 0.2622 |
RGB dataset
| dataset | mean(R, G, B) | std(R, G, B) |
|---|---|---|
| CIFAR10(train) | (0.4914, 0.4822, 0.4465) | (0.2022, 0.1993, 0.2009) |
| CIFAR100(train) | (0.5071, 0.4865, 0.4409) | (0.2008, 0.1983, 0.2022) |
| Caltech101(all images) | (0.5487, 0.5313, 0.5050) | (0.2497, 0.2467, 0.2483) |
| Caltech256(all images) | (0.5520, 0.5336, 0.5050) | (0.2420, 0.2412, 0.2438) |
Link
Mean and std calculations are based on https://discuss.pytorch.org/t/about-normalization-using-pre-trained-vgg16-networks/23560/5