import os
import gradio as gr
import torch
import torch.nn.functional as F
import numpy as np
from PIL import Image
from transformers import pipeline

depth_estimator = pipeline(task="depth-estimation", model="Intel/dpt-hybrid-midas")

def launch(input_image):
    out = depth_estimator(input_image)

    predicted_depth = torch.tensor(out["predicted_depth"])
    
    if len(predicted_depth.shape) == 2:  # Если двумерен, добавляем оси
        predicted_depth = predicted_depth.unsqueeze(0).unsqueeze(0)
    
    prediction = F.interpolate(
        predicted_depth,
        size=input_image.size[::-1],  # Порядок: (ширина, высота)
        mode="bicubic",
        align_corners=False,
    )

    output = prediction.squeeze().numpy()
    formatted = (output * 255 / np.max(output)).astype("uint8")
    depth = Image.fromarray(formatted)
    return depth

iface = gr.Interface(
    launch,
    inputs=gr.Image(type="pil"),
    outputs=gr.Image(type="pil"),
)

demo = gr.Blocks()

with demo:
    gr.TabbedInterface(
        [iface],
        ["Depth Estimation Interface"],
    )

demo.launch(debug=True)