import gradio as gr
import tensorflow as tf
import numpy as np
import cv2
import os

# Load model
model = tf.keras.models.load_model('best_model.keras')

# Video processor class
class SingleVideoProcessor:
    def __init__(self, video_path, max_frames=50, frame_size=(128, 128), frame_interval=3):
        self.video_path = video_path
        self.max_frames = max_frames
        self.frame_size = frame_size
        self.frame_interval = frame_interval

    def process(self):
        frames = self.load_video_frames()
        frames = self.resize_frames(frames)
        frames = self.pad_video_frames(frames)
        return np.array(frames, dtype=np.float32)

    def load_video_frames(self):
        frames = []
        cap = cv2.VideoCapture(self.video_path)
        frame_count = 0
        while cap.isOpened():
            ret, frame = cap.read()
            if not ret:
                break
            frame_count += 1
            if frame_count % self.frame_interval == 0:
                frame = cv2.cvtColor(frame, cv2.COLOR_BGR2RGB)
                frames.append(frame)
        cap.release()
        return frames

    def resize_frames(self, frames):
        return [cv2.resize(frame, self.frame_size) for frame in frames]

    def pad_video_frames(self, frames):
        if len(frames) < self.max_frames:
            padding = self.max_frames - len(frames)
            frames.extend([np.zeros_like(frames[0])] * padding)
        else:
            frames = frames[:self.max_frames]
        return frames

# Prediction function
def predict_violence(video_path):
    processor = SingleVideoProcessor(video_path)
    processed_video = processor.process()
    processed_video = np.expand_dims(processed_video, axis=0)
    prediction = model.predict(processed_video)[0][0]

    label = "Violence" if prediction >= 0.5 else "Non-Violence"
    confidence = f"{prediction:.4f}"
    print(prediction)
    return f"{label} (Confidence: {confidence})"

# Example videos (ensure these exist in your working directory)
examples = [
    ["NV_1.mp4"],
    ["V_1000.mp4"],
    ["V_102.mp4"],
    ["NV_11.mp4"]
]

# Launch Gradio Interface
demo = gr.Interface(
    fn=predict_violence,
    inputs=gr.Video(label="Upload or Select a Video"),
    outputs=gr.Text(label="Prediction"),
    examples=examples,
    title="Violence Detection in Video",
    description="This model classifies videos as Violence or Non-Violence. Upload a short clip or select from examples."
)

demo.launch()