gistfile1.txt

from PIL import Image
import depth_pro
import cv2
import torch
import numpy as np
global depth

def click_event(event, x, y, flags, params): 
  
    # checking for left mouse clicks 
    if event == cv2.EVENT_LBUTTONDOWN: 
        print(depth[y,x])

def renormalize_to_255(arr):
    # Find the minimum and maximum of the array
    min_val = np.min(arr)
    max_val = np.max(arr)
    
    # Renormalize to the range [0, 255]
    normalized_arr = (arr - min_val) / (max_val - min_val) * 255
    
    # Convert to unsigned 8-bit integer (0-255)
    normalized_arr = normalized_arr.astype(np.uint8)
    
    return normalized_arr


# Load model and preprocessing transform
device_cuda = torch.device("cuda:0")
model, transform = depth_pro.create_model_and_transforms(device=device_cuda)
model.eval()

# Load and preprocess an image.
cam = cv2.VideoCapture(0)

#image, _, f_px = depth_pro.load_rgb('E:/Projects/3D/ml-depth-pro/data/example.jpg')
#print(f_px)
f_px = None

ret, image = cam.read()
cv2.imshow('Camera', image[:,:,0])
cv2.setMouseCallback('Camera', click_event) 

while True:
    ret, image = cam.read()
    image = cv2.cvtColor(image, cv2.COLOR_BGR2RGB) 
    if ret:
        IMG = transform(image)

        # Run inference.
        prediction = model.infer(IMG, f_px=f_px)
        depth = prediction["depth"].cpu().detach().numpy()
        depth_img = renormalize_to_255(depth)
        cv2.imshow('Camera', depth_img)
        focallength_px = prediction["focallength_px"]  # Focal length in pixels.
        if cv2.waitKey(0) == ord('q'):
            break