app.py

from torchvision import models, transforms
from PIL import Image
import torch
import torch.nn as nn
import io
import streamlit as st
import time

st.title("パーソナルカラー診断AI")

SIZE = 224
MEAN = (0.485, 0.456, 0.406)
STD = (0.229, 0.224, 0.225)

transform = transforms.Compose([
    transforms.Resize((SIZE, SIZE)),
    transforms.ToTensor(),
    transforms.Normalize(MEAN, STD),
])

model = models.resnet152(pretrained=True)
n_classes = 4
num_ftrs = model.fc.in_features
model.fc = nn.Linear(num_ftrs, n_classes)
device = torch.device("cuda" if torch.cuda.is_available() else "cpu")
model.load_state_dict(torch.load('Resnet_2024_0214_version1', map_location=device))
model.to(device)
model.eval()
view_flag = True
skip = False

def predict_image(img):
    img = img.convert('RGB')
    img_transformed = transform(img)
    inputs = img_transformed.unsqueeze(0).to(device)
    with torch.no_grad():
        outputs = model(inputs)
        _, preds = torch.max(outputs, 1)
    return preds.item()
uploaded_file = st.file_uploader('Choose an image...', type=['jpg', 'png'])
if uploaded_file:
    img = Image.open(uploaded_file)
    st.image(img, caption="Uploaded Image", use_column_width=True)
    pred = predict_image(img)
    if pred == 0:
        season_type = "秋"
    elif pred == 1:
        season_type = "春"
    elif pred == 2:
        season_type = "夏"
    else:
        season_type = "冬"
    if 'show_video' not in st.session_state:
        st.session_state.show_video = False
    if 'skip' not in st.session_state:
        st.session_state.skip = False
    if 'result' not in st.session_state:
        st.session_state.result = False

    st.write(f"パーソナルカラー診断結果:{season_type} ")
    st.write("あなたにおすすめの色はこちらです")

    st.session_state.result = True
    st.image(f"{season_type}.png")
    st.write(
        """
        あなたにおすすめの商品はこちらです
        """)
    st.image("服.png")