ChrisMoe/Chinese_handwriting_model

Chinese Handwriting Recognition – HSK 1

A CNN image classifier trained on the CASIA-HWDB dataset to recognise 178 HSK-1 level Chinese characters from handwritten images.

Model Architecture

• Backbone: ResNet-18 (pretrained on ImageNet, fine-tuned)
• Final layer: Dropout(0.4) → Linear(178 classes)
• Input: RGB image resized to 64 × 64

Performance

• Best Validation Accuracy: 98.13%

Training Setup

Hyperparameter	Value
Optimizer	AdamW
Learning rate	1e-3
Weight decay	1e-4
Scheduler	OneCycleLR
Epochs	15
Batch size	128
GPUs	T4 x2
Augmentation	RandomRotation(10°), RandomAffine, ColorJitter

Usage

import torch
from PIL import Image
from torchvision import transforms
import torch.nn as nn
import torchvision.models as models
from huggingface_hub import hf_hub_download

# ── Download checkpoint ───────────────────────────────
ckpt_path  = hf_hub_download('ChrisMoe/Chinese_handwriting_model', 'chinese_hsk1_model.pth')
checkpoint = torch.load(ckpt_path, map_location='cpu')

# ── Rebuild model ─────────────────────────────────────
class ChineseCharCNN(nn.Module):
    def __init__(self, num_classes, dropout=0.4):
        super().__init__()
        backbone = models.resnet18(weights=None)
        in_features = backbone.fc.in_features
        backbone.fc = nn.Sequential(
            nn.Dropout(dropout),
            nn.Linear(in_features, num_classes)
        )
        self.model = backbone
    def forward(self, x):
        return self.model(x)

model = ChineseCharCNN(num_classes=checkpoint['num_classes'])
model.load_state_dict(checkpoint['model_state_dict'])
model.eval()

idx2char = checkpoint['idx2char']

# ── Inference ─────────────────────────────────────────
transform = transforms.Compose([
    transforms.Resize((64, 64)),
    transforms.ToTensor(),
    transforms.Normalize([0.5, 0.5, 0.5], [0.5, 0.5, 0.5])
])
img    = Image.open('your_character.png').convert('RGB')
tensor = transform(img).unsqueeze(0)
with torch.no_grad():
    logits = model(tensor)
pred_idx = logits.argmax(1).item()
print('Predicted character:', idx2char[pred_idx])