app.py

import cv2
import torch
import numpy as np
from transformers import DPTForDepthEstimation, DPTImageProcessor
import gradio as gr
import torch.nn.utils.prune as prune
import open3d as o3d

device = torch.device("cuda" if torch.cuda.is_available() else "cpu")

model = DPTForDepthEstimation.from_pretrained("Intel/dpt-swinv2-tiny-256", torch_dtype=torch.float32)
model.eval()

# Apply global unstructured pruning
parameters_to_prune = [
    (module, "weight") for module in filter(lambda m: isinstance(m, (torch.nn.Conv2d, torch.nn.Linear)), model.modules())
]
prune.global_unstructured(
    parameters_to_prune,
    pruning_method=prune.L1Unstructured,
    amount=0.4,  # Prune 40% of weights
)

for module, _ in parameters_to_prune:
    prune.remove(module, "weight")

model = torch.quantization.quantize_dynamic(
    model, {torch.nn.Linear, torch.nn.Conv2d}, dtype=torch.qint8
)

model = model.to(device)

processor = DPTImageProcessor.from_pretrained("Intel/dpt-swinv2-tiny-256")

color_map = cv2.applyColorMap(np.arange(256, dtype=np.uint8), cv2.COLORMAP_INFERNO)
color_map = torch.from_numpy(color_map).to(device)

def preprocess_image(image):
    image = cv2.resize(image, (128, 128))
    image = torch.from_numpy(image).permute(2, 0, 1).unsqueeze(0).float().to(device)
    return image / 255.0

def create_point_cloud(depth_map, color_image):
    rows, cols = depth_map.shape
    c, r = np.meshgrid(np.arange(cols), np.arange(rows), sparse=True)
    valid = (depth_map > 0) & (depth_map < 1000)
    z = np.where(valid, depth_map, 0)
    x = np.where(valid, z * (c - cols / 2) / cols, 0)
    y = np.where(valid, z * (r - rows / 2) / rows, 0)
    
    points = np.dstack((x, y, z)).reshape(-1, 3)
    colors = color_image.reshape(-1, 3)
    
    pcd = o3d.geometry.PointCloud()
    pcd.points = o3d.utility.Vector3dVector(points)
    pcd.colors = o3d.utility.Vector3dVector(colors / 255.0)
    
    return pcd

@torch.inference_mode()
def process_frame(image):
    if image is None:
        return None
    preprocessed = preprocess_image(image)
    predicted_depth = model(preprocessed).predicted_depth
    depth_map = predicted_depth.squeeze().cpu().numpy()
    
    # Normalize depth map
    depth_map = (depth_map - depth_map.min()) / (depth_map.max() - depth_map.min())
    
    # Create point cloud
    pcd = create_point_cloud(depth_map, image)
    
    # Visualize point cloud
    vis = o3d.visualization.Visualizer()
    vis.create_window()
    vis.add_geometry(pcd)
    vis.poll_events()
    vis.update_renderer()
    
    # Capture the visualization as an image
    image = vis.capture_screen_float_buffer(False)
    vis.destroy_window()
    
    # Convert the image to numpy array
    point_cloud_image = (np.asarray(image) * 255).astype(np.uint8)
    
    return point_cloud_image

interface = gr.Interface(
    fn=process_frame,
    inputs=gr.Image(sources="webcam", streaming=True),
    outputs="image",
    live=True
)

interface.launch()