app.py

import gradio as gr
import requests
import torch
import torch.nn as nn
from PIL import Image
from torchvision.models import resnet18
from torchvision.transforms import functional as F

def main():
    # モデルの選択
    plant_models = {
        'cucumber':{
            'model_path':'cucumber_resnet18_last_model.pth',
            'labels_list' : ["健全","うどんこ病","灰色かび病","炭疽病","べと病","褐斑病","つる枯病","斑点細菌病","CCYV","モザイク病","MYSV"]
        },
        'eggplant':{
            'model_path':'eggplant_resnet18_last_model.pth',
            'labels_list' : ["健全","うどんこ病","灰色かび病","褐色円星病","すすかび病","半身萎凋病","青枯病"]
        },
        'strawberry':{
            'model_path':'strawberry_resnet18_last_model.pth',
            'labels_list' : ["健全","うどんこ病","炭疽病","萎黄病"]
        },
        'tomato':{
            'model_path':'tomato_resnet18_last_model.pth',
            'labels_list' : ["健全","うどんこ病","灰色かび病","すすかび病","葉かび病","疫病","褐色輪紋病","青枯病","かいよう病","黄化葉巻病"]
        },

    }
    # examples_images=[
    #         ['image/231305_20200302150233_01.JPG'],
    #         ['image/0004_20181120084837_01.jpg'],
    #         ['image/160001_20170830173740_01.JPG'],
    #         ['image/152300_20190119175054_01.JPG'],
            
    #     ]
    # モデルの準備する関数を定義
    def select_model(plant_name):
 
        model_ft = resnet18(num_classes = len(plant_models[plant_name]['labels_list']),pretrained=False)
        device = torch.device("cuda:0" if torch.cuda.is_available() else "cpu")
        model_ft = model_ft.to(device)
        if torch.cuda.is_available():
            model_ft.load_state_dict(torch.load(plant_models[plant_name]['model_path']))
        else:
            model_ft.load_state_dict(
                torch.load(plant_models[plant_name]['model_path'], map_location=torch.device("cpu"))
            )
        model_ft.eval()

        return model_ft

    # 画像分類を行う関数を定義
    @torch.no_grad()
    def inference(gr_input,gr_model_type):
        img = Image.fromarray(gr_input.astype("uint8"), "RGB")

        # 前処理
        img = F.resize(img, (224, 224))
        img = F.to_tensor(img)
        img = img.unsqueeze(0)

        # モデル選択
        model_ft = select_model(gr_model_type)
        # 推論
        output = model_ft(img).squeeze(0)
        probs = nn.functional.softmax(output, dim=0).numpy()
        labels_lenght =len(plant_models[gr_model_type]['labels_list'])
        # ラベルごとの確率をdictとして返す
        return {plant_models[gr_model_type]['labels_list'][i]: float(probs[i]) for i in range(labels_lenght)}
    
    model_labels = list(plant_models.keys())
    
    # 入力の形式を画像とする
    inputs = gr.inputs.Image()
    # モデルの種類を選択する
    model_type = gr.inputs.Radio(model_labels, type='value', label='BASE MODEL')

    # 出力はラベル形式で，top4まで表示する
    outputs = gr.outputs.Label(num_top_classes=4)

    # サーバーの立ち上げ
    interface = gr.Interface(fn=inference, 
        inputs=[inputs, model_type],
        outputs=outputs,
        title='Plant Diseases Diagnosis',
        )

    interface.launch()


if __name__ == "__main__":
    main()