app.py

import numpy as np
import gradio as gr
import cv2
import os
import argparse
from inference import Predictor
import io
#from black import to_black

# os.system("wget https://huggingface.co/YANGYYYY/cartoonize/tree/main/GeneratorV2_train_photo_Hayao_init.pt")
# if os.path.exists("GeneratorV2_train_photo_Hayao_init.pt"):
#     print("下载成功！")
# else:
#     print("下载失败！")

def parse_args():
    parser = argparse.ArgumentParser()
    parser.add_argument('--weight', type=str, default='GeneratorV2_train_photo_Hayao_init.pt')
    parser.add_argument('--device', type=str, default='cpu', help='Device, cuda or cpu')

    return parser.parse_args()

def parse_args_video():
    parser = argparse.ArgumentParser()
    parser.add_argument('--weight', type=str, default='GeneratorV2_train_photo_Hayao_init.pt')
    parser.add_argument('--src', type=str, default='dataset/video/花.mp4', help='Path to input video')
    parser.add_argument('--out', type=str, default='dataset/video_Hayao/hua_hayao.mp4', help='Path to save new video')
    parser.add_argument('--batch-size', type=int, default=4)
    parser.add_argument('--start', type=int, default=0, help='Start time of video (second)')
    parser.add_argument('--end', type=int, default=10, help='End time of video (second), 0 if not set')

    return parser.parse_args()

def transfer(image, transfer_style):
    if transfer_style == "Hayao":
         #output = cv2.cvtColor(image, cv2.COLOR_BGR2GRAY)# 转换为灰度图像
        #os.system("wget https://huggingface.co/YANGYYYY/cartoonize/resolve/main/GeneratorV2_train_photo_Hayao_init.pt")
        args = parse_args()
        predictor = Predictor(args.weight, args.device)
        anime_img = predictor.transform_image(image)
        return anime_img
    elif transfer_style == "Shinkai":
        args = parse_args()
        args.weight = 'GeneratorV2_train_photo_Shinkai_init.pt'
        predictor = Predictor(args.weight, args.device)
        anime_img = predictor.transform_image(image)
        return anime_img
    elif transfer_style == "Kon Satoshi":
        args = parse_args()
        args.weight = 'GeneratorV2_train_photo_Paprika_init.pt'
        predictor = Predictor(args.weight, args.device)
        anime_img = predictor.transform_image(image)
        return anime_img
    else:
        return image

        
def transfer_video(video_input, transfer_style):
    if transfer_style == "Hayao":
         #output = cv2.cvtColor(image, cv2.COLOR_BGR2GRAY)# 转换为灰度图像
        #os.system("wget https://huggingface.co/YANGYYYY/cartoonize/resolve/main/GeneratorV2_train_photo_Hayao_init.pt")
        args = parse_args_video()
        # # 加载视频文件
        # #video_binary = io.BytesIO(video)
            
        # cap = cv2.VideoCapture(video)
        
        # # 读取视频帧并保存到一个列表中
        # video_frames = []
        # while True:
        #     ret, frame = cap.read()
        #     if not ret:
        #         break
        #     video_frames.append(frame)
        
        # # 关闭视频文件
        # cap.release()
        args.src = video_input
        args.out = "video.mp4"
        Predictor(args.weight).transform_video(args.src, args.out, args.batch_size, start=args.start, end=args.end)
        return args.out
        #anime_video = Predictor(args.weight).transform_video(video, args.batch_size, args.start, args.end)
        #return anime_video
    elif transfer_style == "Shinkai":
        args = parse_args()
        args.weight = 'GeneratorV2_train_photo_Shinkai_init.pt'
        predictor = Predictor(args.weight, args.device)
        anime_img = predictor.transform_image(image)
        return anime_img
    elif transfer_style == "Kon Satoshi":
        args = parse_args()
        args.weight = 'GeneratorV2_train_photo_Paprika_init.pt'
        predictor = Predictor(args.weight, args.device)
        anime_img = predictor.transform_image(image)
        return anime_img
    else:
        return image

def clear_output(input_widget):
    input_widget = np.array([])

with gr.Blocks() as demo:
    gr.Markdown("Transfer image or video files using this demo.")
    with gr.Tabs():
        with gr.TabItem("Transfer Image"):
            with gr.Row():
                image_input = gr.Image()
                image_output = gr.Image()
            with gr.Row():
                image_dropdown = gr.Dropdown(label="Transfer Style",choices=["Hayao", "Shinkai", "Kon Satoshi"])
                image_button = gr.Button("Transfer")
                clear_image_button = gr.Button("Clear")
        with gr.TabItem("Transfer Video"):
            with gr.Row():
                video_input = gr.Video()
                video_output = gr.Video()
            with gr.Row():
                video_dropdown = gr.Dropdown(label="Transfer Style",choices=["Hayao", "Shinkai", "Kon Satoshi"])
                video_button = gr.Button("Transfer")
                clear_video_button = gr.Button("Clear")

    image_button.click(transfer, inputs=[image_input,image_dropdown], outputs=image_output)
    video_button.click(transfer_video, inputs=[video_input,video_dropdown],outputs=video_output)
    clear_image_button.click(clear_output, inputs=image_input,outputs=image_output)
    clear_video_button.click(clear_output, inputs=video_input,outputs=video_output)
demo.launch()

# 启动接口
#demo.launch(server_name='127.0.0.1',server_port=7788)