• Introduction
• Prerequisites
• Training

The CIFAR-10 dataset consists of 60000 32x32 colour images in 10 classes, with 6000 images per class. There are 50000 training images and 10000 test images.

below is the 6 random images with their respective label:

There is a package of python called torchvision, that has data loaders for CIFAR10 and data transformers for images using torch.utils.data.DataLoader.

Below an example of how to load CIFAR10 dataset using torchvision:

import torch
import torchvision
## load data CIFAR10
train_dataset = torchvision.datasets.CIFAR10(root='./train_data', train=True, download=True)
train_dataloader = torch.utils.data.DataLoader(train_dataset, batch_size=128, shuffle=True, num_workers=2)

• Python>=3.6
• PyTorch >=1.4
• Library are mentioned in requirenments.txt

I used pretrained resnet18 for model training. you can use any other pretrained model according to you problem.

import torchvision.models as models
alexnet = models.alexnet()
vgg16 = models.vgg16()
densenet = models.densenet161()
inception = models.inception_v3()

There are two things for pytorch model training:

1. Notebook - you can just download and play with it
2. python scripts:

   # Start training with: 
   python main.py
   
   # You can manually pass the attributes for the training: 
   python main.py --lr=0.01 --epoch 20 --model_path './cifar_model.pth'
   
   # Start infrence with:
    python3.6 prediction.py --model_path './cifar_model.pth'