app.py

import torch
from torch import nn
import torch.nn.functional as F
from torchvision.transforms import ToTensor

# Define model
class ConvNet(nn.Module):
    def __init__(self):
        super(ConvNet, self).__init__()
        self.conv1 = nn.Conv2d(1, 32, kernel_size=5)
        self.conv2 = nn.Conv2d(32, 32, kernel_size=5)
        self.conv3 = nn.Conv2d(32,64, kernel_size=5)
        self.fc1 = nn.Linear(3*3*64, 256)
        self.fc2 = nn.Linear(256, 10)

    def forward(self, x):
        x = F.relu(self.conv1(x))
        #x = F.dropout(x, p=0.5, training=self.training)
        x = F.relu(F.max_pool2d(self.conv2(x), 2))
        x = F.dropout(x, p=0.5, training=self.training)
        x = F.relu(F.max_pool2d(self.conv3(x),2))
        x = F.dropout(x, p=0.5, training=self.training)
        x = x.view(-1,3*3*64 )
        x = F.relu(self.fc1(x))
        x = F.dropout(x, training=self.training)
        logits = self.fc2(x)
        return logits


model = ConvNet()
model.load_state_dict(
    torch.load("weights/mnist_convnet_model.pth",
    map_location=torch.device('cpu'))
    )

model.eval()

import gradio as gr
from torchvision import transforms

import os
import glob

examples_dir = './examples'
example_files = glob.glob(os.path.join(examples_dir, '*.png'))

def predict(image):
    tsr_image = transforms.ToTensor()(image)

    with torch.no_grad():
        pred = model(tsr_image)
        prob = torch.nn.functional.softmax(pred[0], dim=0)

    confidences = {i: float(prob[i]) for i in range(10)} 
    return confidences


with gr.Blocks(css=".gradio-container {background:honeydew;}", title="MNIST 分類器"
               ) as demo:
    gr.HTML("""<div style="font-family:'Times New Roman', 'Serif'; font-size:16pt; font-weight:bold; text-align:center; color:royalblue;">MNIST 分類器</div>""")

    with gr.Row():
        with gr.Tab("キャンバス"):
            input_image1 = gr.Image(label="スケッチ", source="canvas", type="pil", image_mode="L", shape=(28,28), invert_colors=True)    
            send_btn1 = gr.Button("推論する")

        with gr.Tab("画像ファイル"):
            input_image2 = gr.Image(label="画像入力", type="pil", image_mode="L", shape=(28, 28), invert_colors=True)
            send_btn2 = gr.Button("推論する")

            gr.Examples(example_files, inputs=input_image2)
            #gr.Examples(['examples/sample02.png', 'examples/sample04.png'], inputs=input_image2)

        output_label=gr.Label(label="推論確率", num_top_classes=3)

    send_btn1.click(fn=predict, inputs=input_image1, outputs=output_label)
    send_btn2.click(fn=predict, inputs=input_image2, outputs=output_label)

# demo.queue(concurrency_count=3)
demo.launch()


### EOF ###