app.py

from PIL import Image
import cv2
import numpy as np
import torch
import torch.nn.functional as F
from torchvision.transforms import Compose
import shutil
import os


from depthAnything.depth_anything.dpt import DepthAnything
from depthAnything.depth_anything.util.transform import Resize, NormalizeImage, PrepareForNet
from TEED.main import parse_args, main 

# 深度处理函数
def depth_anything_image(image, encoder='vitl', pred_only=True, grayscale=True):
    DEVICE = 'cuda' if torch.cuda.is_available() else 'cpu'
    model_configs = {
        'vitl': {'encoder': 'vitl', 'features': 256, 'out_channels': [256, 512, 1024, 1024]},
        'vitb': {'encoder': 'vitb', 'features': 128, 'out_channels': [96, 192, 384, 768]},
        'vits': {'encoder': 'vits', 'features': 64, 'out_channels': [48, 96, 192, 384]}
    }

    depth_anything = DepthAnything(model_configs[encoder])
    depth_anything.load_state_dict(torch.load(f'./checkpoints/depth_anything_{encoder}14.pth'))
    depth_anything = depth_anything.to(DEVICE).eval()

    transform = Compose([
        Resize(width=518, height=518, resize_target=False, keep_aspect_ratio=True,
               ensure_multiple_of=14, resize_method='lower_bound', image_interpolation_method=cv2.INTER_CUBIC),
        NormalizeImage(mean=[0.485, 0.456, 0.406], std=[0.229, 0.224, 0.225]),
        PrepareForNet(),
    ])

    raw_image = np.array(image.convert('RGB'))[:, :, ::-1].copy()  # RGB to BGR
    h, w = raw_image.shape[:2]
    image_tensor = transform({'image': raw_image / 255.0})['image']
    image_tensor = torch.from_numpy(image_tensor).unsqueeze(0).to(DEVICE)

    with torch.no_grad():
        depth = depth_anything(image_tensor)

    depth = F.interpolate(depth[None], (h, w), mode='bilinear', align_corners=False)[0, 0]
    depth = (depth - depth.min()) / (depth.max() - depth.min()) * 255.0
    depth = depth.cpu().numpy().astype(np.uint8)

    return np.repeat(depth[..., np.newaxis], 3, axis=-1) if grayscale else cv2.applyColorMap(depth, cv2.COLORMAP_INFERNO)

# TEED 图像处理函数
def teed_process_image(image):
    os.makedirs('./output/teed_imgs', exist_ok=True)
    os.makedirs('./teed_tmp', exist_ok=True)

    temp_image_path = './teed_tmp/temp_image.png'
    cv2.imwrite(temp_image_path, np.array(image))

    args, train_info = parse_args(is_testing=True, pl_opt_dir='./output/teed_imgs')
    args.input_val_dir = './teed_tmp'
    args.output_dir = './output/teed_imgs'

    checkpoint_path =  './TEED/checkpoints/BIPED/5/5_model.pth'
    if not os.path.exists(checkpoint_path):
        raise FileNotFoundError(f"Checkpoint file not found: {checkpoint_path}")
    
    args.checkpoint_data = checkpoint_path  # 确保使用正确的路径

    main(args, train_info)

    shutil.rmtree('./teed_tmp')
    return cv2.imread(os.path.join('./output/teed_imgs', 'processed_image.png'))



# 处理单个图像
def process_single_image(image):
    depth_result = depth_anything_image(image, 'vitl')
    teed_result = teed_process_image(image)
    merged_result = multiply_blend(depth_result, teed_result)
    return merged_result

# Gradio 界面处理函数
def gradio_process_line(img):
    processed_image = process_single_image(img)
    return Image.fromarray(processed_image)

# Gradio 界面
import gradio as gr

iface = gr.Interface(
    fn=gradio_process_line,
    inputs=gr.Image(type="pil"),
    outputs=gr.Image(type="pil"),
    title="Image Processing with Depth Anything and TEED",
    description="Upload an image to process it with depth estimation and edge detection."
)

# 启动 Gradio 应用
iface.launch(share=True)