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


LABELS = Path('classes.txt').read_text().splitlines()
num_classes = len(LABELS)

model = nn.Sequential(
    nn.Conv2d(1, 64, 3, padding='same'),
    nn.ReLU(),
    nn.MaxPool2d(2),
    nn.Conv2d(64, 128, 3, padding='same'),
    nn.ReLU(),
    nn.MaxPool2d(2),
    nn.Conv2d(128, 256, 3, padding='same'),
    nn.ReLU(),
    nn.MaxPool2d(2),
    nn.Flatten(),
    nn.Linear(2304, 512),
    nn.ReLU(),
    nn.Linear(512, num_classes),
)

state_dict = torch.load('model.pth', map_location='cpu')
model.load_state_dict(state_dict, strict=False)
model.eval()

def predict(image):
    image = image['composite']
    image = transforms.Resize((28, 28))(image)
    print(image)
    
    x = torch.tensor(im, dtype=torch.float32).unsqueeze(0).unsqueeze(0) / 255.

    with torch.no_grad():
        out = model(x)

    probabilities = F.softmax(out[0], dim=0)

    values, indices = torch.topk(probabilities, 5)

    return {LABELS[i]: v.item() for i, v in zip(indices, values)}


interface = gr.Interface(predict, inputs='sketchpad', outputs='label', live=True)
interface.launch(debug=True)