gemini_video_analysis.py

import cv2
import gradio as gr
import google.generativeai as genai
import os
import PIL.Image

YOUR_API_KEY= "AIzaSyBjb6LLerzZE6JIIE0YBK6Wn0hqdO9E1Zk"
genai.configure(api_key="AIzaSyBjb6LLerzZE6JIIE0YBK6Wn0hqdO9E1Zk")

# Define the Generative AI model
model = genai.GenerativeModel('gemini-pro-vision')

# Function to capture frames from a video
def frame_capture(video_path, num_frames=5):
    vidObj = cv2.VideoCapture(video_path)
    frames = []
    count = 0
    while True:
        success, image = vidObj.read()
        if not success:
            break
        if count % (vidObj.get(cv2.CAP_PROP_FRAME_COUNT) // num_frames) == 0:
            frames.append(image)
        count += 1
    return frames


#format description

def format_descriptions(descriptions):
    # Join the descriptions into a single string
    formatted_description = ' '.join(descriptions)
    # Remove any leading or trailing whitespace
    formatted_description = formatted_description.strip()
    # Replace any occurrences of special characters with a space
    formatted_description = ''.join(char if char.isalnum() or char.isspace() else ' ' for char in formatted_description)
    return formatted_description

# Function to generate text descriptions for frames
def generate_descriptions_for_frames(video_path):
    # Capture frames from the video
    frames = frame_capture(video_path)
    
    # Prepare images for input to the model
    images = [PIL.Image.fromarray(cv2.cvtColor(frame, cv2.COLOR_BGR2RGB)) for frame in frames]
    
    # Prepare the prompt with images and instructions
    instructions = "Instructions: Consider the following frames from the video:"
    prompt = "What is shown in each of the frames?"
    images_with_prompt = [prompt] + [instructions] + images
    
    # Generate content using the model
    responses = model.generate_content(images_with_prompt)
    
    # Extract and return the text descriptions from the responses
    descriptions = [response.text for response in responses]
    formatted_description = format_descriptions(descriptions)
    
    return formatted_description


# Define Gradio interface
video_input = gr.Video(label="Upload Video", autoplay=True)
output_text = gr.Textbox(label="Frame Descriptions")

# Create Gradio app
gr.Interface(fn= generate_descriptions_for_frames, inputs=video_input, outputs=output_text, title="Frame Description Generator").launch()