app.py

import gradio as gr
import numpy as np
import torch
import torchvision
from torch import nn

class LeNet(nn.Module):
    def __init__(self):
        super(LeNet, self).__init__()

        self.convs = nn.Sequential(
            nn.Conv2d(in_channels=1, out_channels=4, kernel_size=(5, 5)),
            nn.Tanh(),
            nn.AvgPool2d(2, 2),

            nn.Conv2d(in_channels=4, out_channels=12, kernel_size=(5, 5)),
            nn.Tanh(),
            nn.AvgPool2d(2, 2)
        )

        self.linear = nn.Sequential(
            nn.Linear(4*4*12,10)
        )

    def forward(self, x):
        x = self.convs(x)
        x = torch.flatten(x, 1)

        return self.linear(x)

    @torch.no_grad()
    def predict(self, input):
        input = input.reshape(1, 1, 28, 28)
        out = self(input)
        return nn.functional.softmax(out[0], dim = 0)

lenet = LeNet()
lenet.load_state_dict(torch.load('ibob-lenet-v1/lenet-v1.pth', map_location='cpu'))


resize = torchvision.transforms.Resize((28, 28), antialias=True)
def on_submit(img):
    with torch.no_grad():
        img = img['composite'].astype(np.float32)
        img = torch.from_numpy(img)
        img = resize(img.unsqueeze(0))

        result = lenet.predict(img)

        sorted = [[i, e] for i, e in enumerate(result.numpy())]
        sorted.sort(key = lambda a : -a[1])

        return "\n".join(map(str, sorted))

iface = gr.Interface(
    title = "LeNet",
    fn = on_submit,
    inputs=gr.Sketchpad(image_mode='P'),
    outputs=gr.Text(),
)
iface.launch()