app.py

import streamlit as st
st.set_page_config(page_title="Object Volume Estimator", layout="wide")

import cv2
import torch
import numpy as np
from PIL import Image
import pandas as pd
from ultralytics import YOLO
from torchvision.transforms import Compose, Resize, ToTensor, Normalize

# Load models
@st.cache_resource
def load_models():
    yolo = YOLO("yolov8n.pt")
    midas = torch.hub.load("intel-isl/MiDaS", "DPT_Large")
    midas.eval()
    transform = Compose([
        Resize(384),
        ToTensor(),
        Normalize(mean=[0.5, 0.5, 0.5], std=[0.5, 0.5, 0.5])
    ])
    return yolo, midas, transform

yolo_model, midas_model, midas_transform = load_models()

# --- Streamlit App ---
st.title("Object Dimension & Volume Estimator")

# 📸 Image Source Selection
option = st.radio("Choose Image Source:", ("📤 Upload Image", "📸 Use Camera"))

# Mode Selection
mode = st.selectbox("Select Image Type:", ["2D RGB Image", "RGB + Depth Image"])

image_pil = None

if option == "📤 Upload Image":
    uploaded_file = st.file_uploader("Upload an image", type=["jpg", "jpeg", "png"])
    if uploaded_file:
        image_pil = Image.open(uploaded_file).convert("RGB")

elif option == "📸 Use Camera":
    camera_input = st.camera_input("Take a picture")
    if camera_input:
        image_pil = Image.open(camera_input).convert("RGB")

# Upload depth image if RGB + Depth selected
depth_map = None
if mode == "RGB + Depth Image" and image_pil is not None:
    depth_file = st.file_uploader("Upload corresponding Depth Map (grayscale image)", type=["png", "jpg", "jpeg", "tiff"])
    if depth_file:
        depth_map = np.array(Image.open(depth_file).convert("L"))

# Proceed if we have an image
if image_pil is not None:
    # Convert to OpenCV format
    image_cv = np.array(image_pil)
    image_cv = cv2.cvtColor(image_cv, cv2.COLOR_RGB2BGR)
    img_h, img_w = image_cv.shape[:2]
    st.image(image_pil, caption="Input Image", use_container_width=True)

    # Store detected object data
    object_data = []

    # 🔍 YOLO Object Detection
    results = yolo_model(image_cv)

    # 📏 Depth Estimation
    if mode == "2D RGB Image":
        input_tensor = midas_transform(image_pil).unsqueeze(0)
        with torch.no_grad():
            estimated_depth = midas_model(input_tensor).squeeze().cpu().numpy()
        depth_map_resized = cv2.resize(estimated_depth, (img_w, img_h))
    else:
        if depth_map is not None:
            depth_map_resized = cv2.resize(depth_map, (img_w, img_h))
        else:
            st.error("Please upload a corresponding depth map for 3D images.")
            st.stop()

    # 🎯 Process each detection
    for r in results:
        for box in r.boxes:
            x1, y1, x2, y2 = map(int, box.xyxy[0])
            width = x2 - x1
            height = y2 - y1
            region = depth_map_resized[y1:y2, x1:x2]
            if region.size == 0:
                continue
            depth = np.mean(region)
            volume = round(depth * width * height, 2)

            # Draw bounding box & label
            cv2.rectangle(image_cv, (x1, y1), (x2, y2), (0, 255, 0), 2)
            label = f"L×B×H: {depth:.2f}×{width}×{height} | V: {volume}"
            cv2.putText(image_cv, label, (x1, y1 - 10),
                        cv2.FONT_HERSHEY_SIMPLEX, 0.5, (255, 255, 0), 2)

            # Store data for CSV
            object_data.append({
                "Length": round(depth, 2),
                "Breadth": int(width),
                "Height": int(height),
                "Volume": volume
            })

            # Show object details
            st.markdown(f"**🧠 Object Detected:**")
            st.write(f"📏 Length × Breadth × Height: {depth:.2f} × {width} × {height}")
            st.write(f"📦 Estimated Volume: {volume} (relative units³)")

    # Show annotated image
    result_img = cv2.cvtColor(image_cv, cv2.COLOR_BGR2RGB)
    st.image(result_img, caption="Detected Dimensions", use_container_width=True)

    # Show table & CSV download if data exists
    if object_data:
        df = pd.DataFrame(object_data)
        st.markdown("### 📋 Detected Objects Table")
        st.dataframe(df)

        csv = df.to_csv(index=False).encode('utf-8')
        st.download_button(
            label="📥 Download Results as CSV",
            data=csv,
            file_name='object_dimensions_volume.csv',
            mime='text/csv',
        )