app.py

import streamlit as st
import cv2
import torch
import numpy as np
from ultralytics import YOLO
from streamlit_webrtc import webrtc_streamer, VideoTransformerBase

# Import local UniDepth
from unidepth.models import UniDepthV2

# -----------------------------
# Streamlit setup
# -----------------------------
st.set_page_config(page_title="YOLO + UniDepth Streaming", layout="wide")
st.title("📡 YOLO + UniDepth V2 — Real 3D Distance (CPU Edition)")

# -----------------------------
# Load models (cached)
# -----------------------------
@st.cache_resource
def load_depth_model():
    device = torch.device("cuda" if torch.cuda.is_available() else "cpu")
    model = UniDepthV2.from_pretrained("lpiccinelli/unidepth-v2-vitl14").to(device)
    model.eval()
    return model, device

@st.cache_resource
def load_yolo():
    return YOLO("yolov8n.pt")

depth_model, device = load_depth_model()
yolo_model = load_yolo()

# -----------------------------
# Helper functions
# -----------------------------
def get_depth_points(rgb):
    img_t = torch.from_numpy(rgb).float().permute(2, 0, 1).unsqueeze(0).to(device)/255.0
    with torch.no_grad():
        pred = depth_model.infer(img_t)
    return pred["points"][0].cpu().numpy()

def process_frame(frame_bgr):
    frame_rgb = cv2.cvtColor(frame_bgr, cv2.COLOR_BGR2RGB)
    points_3d = get_depth_points(frame_rgb)

    results = yolo_model(frame_bgr, verbose=False)[0]
    for box in results.boxes:
        conf = box.conf.item()
        if conf < 0.5: 
            continue

        cls_id = int(box.cls.item())
        label = yolo_model.names[cls_id]

        x1, y1, x2, y2 = map(int, box.xyxy[0].tolist())
        cx, cy = (x1 + x2)//2, (y1 + y2)//2

        X, Y, Z = points_3d[:, cy, cx]
        distance = np.sqrt(X**2 + Y**2 + Z**2)

        cv2.rectangle(frame_bgr, (x1,y1),(x2,y2),(0,255,0),2)
        cv2.putText(frame_bgr, f"{label} {distance:.2f}m", (x1,y1-10),
                    cv2.FONT_HERSHEY_SIMPLEX, 0.7, (0,255,255), 2)

    return frame_bgr

# -----------------------------
# WebRTC Video Processor
# -----------------------------
class YOLODepthProcessor(VideoTransformerBase):
    def transform(self, frame):
        img = frame.to_ndarray(format="bgr24")
        return process_frame(img)

# -----------------------------
# WebRTC Streamlit Widget
# -----------------------------
st.warning("This app runs in the cloud — your webcam stays private on your device.")

webrtc_streamer(
    key="yolo-depth",
    video_transformer_factory=YOLODepthProcessor,
    async_processing=True,
    media_stream_constraints={"video": True, "audio": False},
    rtc_configuration={
        "iceServers":[
            {"urls":["stun:stun.l.google.com:19302"]},
            {
                "urls":["turn:openrelay.metered.ca:80"],
                "username":"openrelayproject",
                "credential":"openrelayproject"
            }
        ]
    }
)