app.py

import numpy as np
import torch
from pathlib import Path
import torch.nn as nn
import torch.nn.functional as F
from PIL import Image
from torchvision import transforms
import gradio as gr


transform = transforms.Compose([
    transforms.Resize((28, 28)),
    transforms.Grayscale(),
    transforms.ToTensor()
])
labels = ["๐ (ศูนย์)", "๑ (หนึ่ง)", "๒ (สอง)", "๓ (สาม)", "๔ (สี่)", "๕ (ห้า)", "๖ (หก)", "๗ (เจ็ด)", "๘ (แปด)", "๙ (เก้า)"]
LABELS = {i:k for i, k in enumerate(labels)} # dictionary of index and label


# Load model using DropoutThaiDigit instead
class DropoutThaiDigit(nn.Module):
    def __init__(self):
        super(DropoutThaiDigit, self).__init__()
        self.fc1 = nn.Linear(28 * 28, 392)
        self.fc2 = nn.Linear(392, 196)
        self.fc3 = nn.Linear(196, 98)
        self.fc4 = nn.Linear(98, 10)
        self.dropout = nn.Dropout(0.1)
        
    def forward(self, x):
        x = x.view(-1, 28 * 28)
        x = self.fc1(x)
        x = F.relu(x)
        x = self.dropout(x)
        x = self.fc2(x)
        x = F.relu(x)
        x = self.dropout(x)
        x = self.fc3(x)
        x = F.relu(x)
        x = self.dropout(x)
        x = self.fc4(x)
        return x


model = DropoutThaiDigit()
model.load_state_dict(torch.load("thai_digit_net.pth"))
model.eval()


def predict(img):
    """
    Predict function takes image and return top 5 predictions
    as a dictionary:
        {label: confidence, label: confidence, ...}
    """
    if img.get("composite") is not None:
        if img["composite"].sum() == 0:
            return {"No input sketch": 0.0}

    img_data = img['composite']
    img_gray = Image.fromarray(img_data).convert('L').resize((28, 28))
    img_tensor = transforms.ToTensor()(img_gray).unsqueeze(0)

    # Make prediction
    with torch.no_grad():
        probs = model(img_tensor).softmax(dim=1).squeeze()

    probs, indices = torch.topk(probs, 5)  # select top 5
    probs, indices = probs.tolist(), indices.tolist()  # transform to list
    return {LABELS[i]: float(v) for i, v in zip(indices, probs)}

js_func = """
function refresh() {
    const url = new URL(window.location);

    if (url.searchParams.get('__theme') !== 'dark') {
        url.searchParams.set('__theme', 'dark');
        window.location.href = url.href;
    }
}
"""

with gr.Blocks(js=js_func) as demo:
    gr.Interface(
        fn=predict,
        inputs=gr.Sketchpad(
            label="Draw Here",
            brush=gr.Brush(default_size=14, default_color="#FFFFFF", colors=["#FFFFFF"]),
            image_mode="L",
            layers=False,
            eraser=None,
            width=400,
            height=350
        ),
        outputs=gr.Label(label="Guess"),
        title="Thai Digit Handwritten Classification",
        description="ทดลองวาดภาพตัวอักษรเลขไทยลงใน Sketchpad ด้านล่างเพื่อทำนายผลตัวเลข ตั้งแต่ ๐ (ศูนย์) ๑ (หนึ่ง) ๒ (สอง) ๓ (สาม) ๔ (สี่) ๕ (ห้า) ๖ (หก) ๗ (เจ็ด) ๘ (แปด) จนถึง ๙ (เก้า)",
        live=True
    )

if __name__ == "__main__":
    demo.launch()