app.py

import depth_pro
import gradio as gr
import matplotlib.cm as cm
import numpy as np
from depth_pro.depth_pro import DepthProConfig
from PIL import Image

MARKDOWN = """
        <div align="center">
            <h2><a href="https://arxiv.org/abs/2410.02073">Depth Pro: Sharp Monocular Metric Depth in Less Than a Second</a></h2>
        </div>
"""


def run(input_image_path):
    config = DepthProConfig(
        patch_encoder_preset="dinov2l16_384",
        image_encoder_preset="dinov2l16_384",
        checkpoint_uri="./depth_pro.pt",
        decoder_features=256,
        use_fov_head=True,
        fov_encoder_preset="dinov2l16_384",
    )

    # Load model and preprocessing transform
    model, transform = depth_pro.create_model_and_transforms(config=config)
    model.eval()

    # Load and preprocess an image
    image, _, f_px = depth_pro.load_rgb(input_image_path)
    image = transform(image)

    # Run inference
    prediction = model.infer(image, f_px=f_px)

    depth_map = prediction["depth"].squeeze().cpu().numpy()
    focallength_px = prediction["focallength_px"]

    depth_map = (depth_map - depth_map.min()) / (depth_map.max() - depth_map.min())
    colormap = cm.get_cmap("viridis")
    depth_map = colormap(depth_map)
    depth_map = (depth_map[:, :, :3] * 255).astype(np.uint8)

    depth_map = Image.fromarray(depth_map)

    return depth_map, focallength_px.item()


with gr.Blocks() as demo:
    gr.Markdown(MARKDOWN)
    with gr.Row():
        with gr.Column():
            input_image_path = gr.Image(
                label="Input Image", type="filepath", sources=["upload"]
            )
        with gr.Column():
            with gr.Column():
                output_depth_map = gr.Image(label="Depth Map")
                output_focal_length = gr.Number(label="Focal Length")

    with gr.Row():
        btn = gr.Button("Run")

    btn.click(
        run, inputs=[input_image_path], outputs=[output_depth_map, output_focal_length]
    )

    examples = gr.Examples(
        examples=[
            "assets/input_one.webp",
        ],
        fn=run,
        inputs=[input_image_path],
        outputs=[output_depth_map, output_focal_length],
        cache_examples=True,
    )

demo.launch()