app.py

import gradio as gr
import cv2
import numpy as np
import os
from PIL import Image
import torch
import torch.nn.functional as F
from torchvision.transforms import Compose
import tempfile
from gradio_imageslider import ImageSlider

from depth_anything.dpt import DepthAnything
from depth_anything.util.transform import Resize, NormalizeImage, PrepareForNet

css = """
#img-display-container {
    max-height: 100vh;
    }
#img-display-input {
    max-height: 80vh;
    }
#img-display-output {
    max-height: 80vh;
    }
"""
DEVICE = 'cpu'
encoder = 'vitl' # can also be 'vitb' or 'vitl'
model = DepthAnything.from_pretrained(f"LiheYoung/depth_anything_{encoder}14").to(DEVICE).eval()

title = "# Depth Anything with log"

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(),
])

@torch.no_grad()
def predict_depth(model, image):
    return model(image)


with (gr.Blocks(css=css) as demo):
    gr.Markdown(title)

    with gr.Row():
        input_image = gr.Image(label="Input Image", type='numpy', elem_id='img-display-input')
        depth_image_slider = ImageSlider(label="Depth Map with Slider View", elem_id='img-display-output', position=0.5,)
    raw_file = gr.File(label="16-bit raw depth (can be considered as disparity)")
    submit = gr.Button("Submit")

    def on_submit(image):
        original_image = image.copy()

        h, w = image.shape[:2]

        image = cv2.cvtColor(image, cv2.COLOR_BGR2RGB) / 255.0
        image = transform({'image': image})['image']
        image = torch.from_numpy(image).unsqueeze(0).to(DEVICE)

        depth = predict_depth(model, image)
        depth = F.interpolate(depth[None], (h, w), mode='bilinear', align_corners=False)[0, 0]

        raw_depth = Image.fromarray(depth.cpu().numpy().astype('uint16'))
        tmp = tempfile.NamedTemporaryFile(suffix='.png', delete=False)
        raw_depth.save(tmp.name)

        # depth = (depth - depth.min()) / (depth.max() - depth.min()) *255.
        image_flattened = depth.view(image.size(0), -1)

        # 计算分位数阈值
        lower_quantile = torch.quantile(image_flattened, 0.05, dim=1, keepdim=True)
        upper_quantile = torch.quantile(image_flattened, 0.95, dim=1, keepdim=True)

        # 应用阈值，去除极值
        clamped_image_flattened = torch.clamp(image_flattened, lower_quantile, upper_quantile)

        # 恢复图像到原始形状
        clamped_image = clamped_image_flattened.view_as(depth)

        epsilon = 1e-7  # 一个小的正值，以避免计算log(0)
        log_image = torch.log(clamped_image + epsilon)
        depth = (log_image - log_image.min()) / (log_image.max() - log_image.min()) *255.

        depth = depth.cpu().numpy().astype(np.uint8)

        return [(original_image, depth), tmp.name]

    submit.click(on_submit, inputs=[input_image], outputs=[depth_image_slider, raw_file])

if __name__ == '__main__':
    demo.queue().launch()