app.py

import gradio as gr
import cv2
import numpy as np
import torch
from ultralytics import SAM, YOLOWorld
import os

# Initialize models
sam_model = SAM("mobile_sam.pt")  # Switch to MobileSAM for faster CPU inference
yolo_model = YOLOWorld("yolov8n-world.pt")  # Nano model for faster detection

def detect_motorcycles(first_frame, prompt="motorcycle"):
    """Detect motorcycles in the first frame using YOLO-World and return bounding boxes."""
    yolo_model.set_classes([prompt])
    results = yolo_model.predict(first_frame, device="cpu", max_det=2)  # Limit to 2 detections
    boxes = []
    for result in results:
        boxes.extend(result.boxes.xyxy.cpu().numpy())
    if len(boxes) > 0:
        boxes = np.vstack(boxes)
    else:
        boxes = np.array([])
    return boxes

def segment_and_highlight_video(video_path, prompt="motorcycle", highlight_color="red"):
    """Segment and highlight motorcycles in a video using SAM 2 and YOLO-World."""
    # Get first frame for detection
    cap = cv2.VideoCapture(video_path)
    ret, first_frame = cap.read()
    if not ret:
        raise ValueError("Could not read first frame from video.")
    # Resize first frame for detection
    first_frame = cv2.resize(first_frame, (320, 180))
    cap.release()

    # Detect boxes in first frame
    boxes = detect_motorcycles(first_frame, prompt)

    if len(boxes) == 0:
        return video_path  # No motorcycles detected, return original

    # Resize boxes to match SAM input resolution (320x180)
    scale_x = 320 / first_frame.shape[1]
    scale_y = 180 / first_frame.shape[0]
    boxes = boxes * [scale_x, scale_y, scale_x, scale_y]

    # Run SAM on video with boxes prompt
    results = sam_model.predict(source=video_path, bboxes=boxes, stream=True, imgsz=320)  # Stream and low resolution

    # Prepare output video
    cap = cv2.VideoCapture(video_path)
    fps = cap.get(cv2.CAP_PROP_FPS)
    width = 320
    height = 180
    output_path = "output.mp4"
    out = cv2.VideoWriter(output_path, cv2.VideoWriter_fourcc(*"mp4v"), fps, (width, height))

    # Color map for highlighting
    color_map = {"red": (0, 0, 255), "green": (0, 255, 0), "blue": (255, 0, 0)}
    highlight_rgb = color_map.get(highlight_color.lower(), (0, 0, 255))

    frame_idx = 0
    for result in results:
        frame = cv2.VideoCapture(video_path).read()[1]
        frame = cv2.resize(frame, (width, height))

        # Get masks for this frame
        if result.masks is not None:
            masks = result.masks.data.cpu().numpy()  # (num_masks, h, w)
            combined_mask = np.any(masks, axis=0).astype(np.uint8) * 255
            mask_colored = np.zeros_like(frame)
            mask_colored[:, :, 0] = combined_mask * highlight_rgb[0]
            mask_colored[:, :, 1] = combined_mask * highlight_rgb[1]
            mask_colored[:, :, 2] = combined_mask * highlight_rgb[2]
            highlighted_frame = cv2.addWeighted(frame, 0.7, mask_colored, 0.3, 0)
        else:
            highlighted_frame = frame

        out.write(highlighted_frame)
        frame_idx += 1

    cap.release()
    out.release()

    return output_path

# Gradio interface
iface = gr.Interface(
    fn=segment_and_highlight_video,
    inputs=[
        gr.Video(label="Upload Video"),
        gr.Textbox(label="Prompt", placeholder="e.g., motorcycle"),
        gr.Dropdown(choices=["red", "green", "blue"], label="Highlight Color")
    ],
    outputs=gr.Video(label="Highlighted Video"),
    title="Video Segmentation with MobileSAM and YOLO-World (CPU)",
    description="Upload a short video (5-10 seconds), specify a text prompt (e.g., 'motorcycle'), and choose a highlight color. Optimized for CPU."
)
iface.launch()