Shapes of images are the same as the training and i use Grayscale both on training and testing. Here’s the code of model training # Import dependencies
import torch
from torch import nn, save, load
from torch.optim import Adam
from torch.utils.data import DataLoader
from RuHandwrittenLetters import RuHandwrittenDataset
Get data
train = RuHandwrittenDataset(‘labels.csv’,‘dataset’)
dataset = DataLoader(train, 64)
Image Classifier Neural Network
class ImageClassifier(nn.Module):
def init(self):
super().init()
self.model = nn.Sequential(
nn.Conv2d(1, 32, (3, 3)),
nn.ReLU(),
nn.Conv2d(32, 64, (3, 3)),
nn.ReLU(),
nn.Conv2d(64, 64, (3, 3)),
nn.ReLU(),
nn.Flatten(),
nn.Linear(64 * (32 - 6) * (32 - 6), 33)
)
def forward(self, x):
return self.model(x)
device = torch.device(“cpu”)
Instance of the neural network, loss, optimizer/
clf = ImageClassifier().to(device)
opt = Adam(clf.parameters(), lr=0.09)
loss_fn = nn.CrossEntropyLoss()
Training flow
if name == “main”:
for epoch in range(15): # train for 10 epochs
for batch in dataset:
X, y = batch
X, y = X.to(device), y.to(device)
yhat = clf(X)
loss = loss_fn(yhat, y)
# Apply backprop
opt.zero_grad()
loss.backward()
opt.step()
print(f"Epoch:{epoch} loss is {loss.item()}")
with open('model_state2.pt', 'wb') as f:
save(clf.state_dict(), f)
and model opener
import torchvision.io
from torch import load
import test2
import torch
from torchvision.transforms.functional import convert_image_dtype
device=‘cpu’
with open(‘model_state.pt’, ‘rb’) as f:
test2.clf.load_state_dict(load(f))
img_tensor=torchvision.io.read_image(‘00_00_00_0000.png’,torchvision.io.ImageReadMode.GRAY)
img_tensor=convert_image_dtype(img_tensor, dtype=torch.float32)
img_tensor.to(device)
print(torch.argmax(test2.clf(img_tensor)))