app.py

import gradio as gr
import torch
from torch import nn
from PIL import Image
from torchvision import transforms
import torchvision.models as models
import torchvision.transforms as transforms

CLASSES = ['guro', 'pigs', 'proofs', 'protyk', 'safe', 'shit']
NUM_CLASSES = len(CLASSES)

device = torch.device('cuda' if torch.cuda.is_available() else 'cpu')

model = models.resnet18(pretrained=True)
model.fc = nn.Linear(model.fc.in_features, NUM_CLASSES)
model.load_state_dict(torch.load('best_model.pth'))
model.to(device)
model.eval()

# Определение трансформаций для изображений
transform = transforms.Compose([
    transforms.Resize(256),
    transforms.CenterCrop(224),
    transforms.ToTensor(),
    transforms.Normalize(mean=[0.485, 0.456, 0.406], std=[0.229, 0.224, 0.225])
])

# Функция для предсказания
def predict(img):
    img = Image.fromarray(img)
    img = transform(img)

    with torch.no_grad():
        outputs = model(img.unsqueeze(0).to(device))
        probabilities = torch.softmax(outputs, dim=1).to('cpu')
        labels = [CLASSES[i] for i in range(len(CLASSES))]
        result = [dict(zip(labels, probabilities.numpy()[0])), dict(zip(labels, probabilities.numpy()[0]))]

    return result[0]

# Интерфейс Gradio
gr.Interface(fn=predict, inputs="image", outputs="label").launch()