|
| 1 | +# Author: Ansh |
| 2 | +import streamlit as st |
| 3 | +import oci |
| 4 | +import base64 |
| 5 | +import cv2 |
| 6 | +from PIL import Image |
| 7 | +from concurrent.futures import ThreadPoolExecutor |
| 8 | +from tqdm import tqdm |
| 9 | + |
| 10 | +# OCI Configuration |
| 11 | +compartmentId = "ocid1.compartment.oc1..XXXXXXXXXXXXXxxxxxxxxxxxxxxxxxxxxxxxxxxx" |
| 12 | +llm_service_endpoint = "https://inference.generativeai.us-chicago-1.oci.oraclecloud.com" |
| 13 | +CONFIG_PROFILE = "DEFAULT" |
| 14 | +visionModel = "meta.llama-3.2-90b-vision-instruct" |
| 15 | +summarizeModel = "cohere.command-r-plus-08-2024" |
| 16 | +config = oci.config.from_file('~/.oci/config', CONFIG_PROFILE) |
| 17 | +llm_client = oci.generative_ai_inference.GenerativeAiInferenceClient( |
| 18 | + config=config, |
| 19 | + service_endpoint=llm_service_endpoint, |
| 20 | + retry_strategy=oci.retry.NoneRetryStrategy(), |
| 21 | + timeout=(10, 240) |
| 22 | + ) |
| 23 | + |
| 24 | +# Functions for Image Analysis |
| 25 | +def encode_image(image_path): |
| 26 | + with open(image_path, "rb") as image_file: |
| 27 | + return base64.b64encode(image_file.read()).decode("utf-8") |
| 28 | + |
| 29 | +# Functions for Video Analysis |
| 30 | +def encode_cv2_image(frame): |
| 31 | + _, buffer = cv2.imencode('.jpg', frame) |
| 32 | + return base64.b64encode(buffer).decode("utf-8") |
| 33 | + |
| 34 | +# Common Functions |
| 35 | +def get_message(encoded_image=None, user_prompt=None): |
| 36 | + content1 = oci.generative_ai_inference.models.TextContent() |
| 37 | + content1.text = user_prompt |
| 38 | + |
| 39 | + message = oci.generative_ai_inference.models.UserMessage() |
| 40 | + message.content = [content1] |
| 41 | + |
| 42 | + if encoded_image: |
| 43 | + content2 = oci.generative_ai_inference.models.ImageContent() |
| 44 | + image_url = oci.generative_ai_inference.models.ImageUrl() |
| 45 | + image_url.url = f"data:image/jpeg;base64,{encoded_image}" |
| 46 | + content2.image_url = image_url |
| 47 | + message.content.append(content2) |
| 48 | + return message |
| 49 | + |
| 50 | +def get_chat_request(encoded_image=None, user_prompt=None): |
| 51 | + chat_request = oci.generative_ai_inference.models.GenericChatRequest() |
| 52 | + chat_request.messages = [get_message(encoded_image, user_prompt)] |
| 53 | + chat_request.api_format = oci.generative_ai_inference.models.BaseChatRequest.API_FORMAT_GENERIC |
| 54 | + chat_request.num_generations = 1 |
| 55 | + chat_request.is_stream = False |
| 56 | + chat_request.max_tokens = 500 |
| 57 | + chat_request.temperature = 0.75 |
| 58 | + chat_request.top_p = 0.7 |
| 59 | + chat_request.top_k = -1 |
| 60 | + chat_request.frequency_penalty = 1.0 |
| 61 | + return chat_request |
| 62 | + |
| 63 | +def cohere_chat_request(encoded_image=None, user_prompt=None): |
| 64 | + print(" i am here") |
| 65 | + chat_request = oci.generative_ai_inference.models.CohereChatRequest() |
| 66 | + chat_request.api_format = oci.generative_ai_inference.models.BaseChatRequest.API_FORMAT_COHERE |
| 67 | + message = get_message(encoded_image, user_prompt) |
| 68 | + chat_request.message = message.content[0].text |
| 69 | + chat_request.is_stream = False |
| 70 | + chat_request.preamble_override = "Make sure you answer only in "+ lang_type |
| 71 | + chat_request.max_tokens = 500 |
| 72 | + chat_request.temperature = 0.75 |
| 73 | + chat_request.top_p = 0.7 |
| 74 | + chat_request.top_k = 0 |
| 75 | + chat_request.frequency_penalty = 1.0 |
| 76 | + return chat_request |
| 77 | + |
| 78 | + |
| 79 | +def get_chat_detail(chat_request,model): |
| 80 | + chat_detail = oci.generative_ai_inference.models.ChatDetails() |
| 81 | + chat_detail.serving_mode = oci.generative_ai_inference.models.OnDemandServingMode(model_id=model) |
| 82 | + chat_detail.compartment_id = compartmentId |
| 83 | + chat_detail.chat_request = chat_request |
| 84 | + return chat_detail |
| 85 | + |
| 86 | +def extract_frames(video_path, interval=1): |
| 87 | + frames = [] |
| 88 | + cap = cv2.VideoCapture(video_path) |
| 89 | + frame_rate = int(cap.get(cv2.CAP_PROP_FPS)) |
| 90 | + success, frame = cap.read() |
| 91 | + count = 0 |
| 92 | + |
| 93 | + while success: |
| 94 | + if count % (frame_rate * interval) == 0: |
| 95 | + frames.append(cv2.cvtColor(frame, cv2.COLOR_BGR2RGB)) |
| 96 | + success, frame = cap.read() |
| 97 | + count += 1 |
| 98 | + cap.release() |
| 99 | + return frames |
| 100 | + |
| 101 | +def process_frame(llm_client, frame, prompt): |
| 102 | + encoded_image = encode_cv2_image(frame) |
| 103 | + try: |
| 104 | + llm_request = get_chat_request(encoded_image, prompt) |
| 105 | + llm_payload = get_chat_detail(llm_request,visionModel) |
| 106 | + llm_response = llm_client.chat(llm_payload) |
| 107 | + return llm_response.data.chat_response.choices[0].message.content[0].text |
| 108 | + except Exception as e: |
| 109 | + return f"Error processing frame: {str(e)}" |
| 110 | + |
| 111 | +def process_frames_parallel(llm_client, frames, prompt): |
| 112 | + with ThreadPoolExecutor() as executor: |
| 113 | + results = list(tqdm( |
| 114 | + executor.map(lambda frame: process_frame(llm_client, frame, prompt), frames), |
| 115 | + total=len(frames), |
| 116 | + desc="Processing frames" |
| 117 | + )) |
| 118 | + return results |
| 119 | + |
| 120 | +def generate_final_summary(llm_client, frame_summaries): |
| 121 | + combined_summaries = "\n".join(frame_summaries) |
| 122 | + final_prompt = ( |
| 123 | + "You are a video content summarizer. Below are summaries of individual frames extracted from a video. " |
| 124 | + "Using these frame summaries, create a cohesive and concise summary that describes the content of the video as a whole. " |
| 125 | + "Focus on providing insights about the overall theme, events, or key details present in the video, and avoid referring to individual frames or images explicitly.\n\n" |
| 126 | + f"{combined_summaries}" |
| 127 | + ) |
| 128 | + try: |
| 129 | + llm_request = cohere_chat_request(user_prompt=final_prompt) |
| 130 | + llm_payload = get_chat_detail(llm_request,summarizeModel) |
| 131 | + llm_response = llm_client.chat(llm_payload) |
| 132 | + return llm_response.data.chat_response.text |
| 133 | + except Exception as e: |
| 134 | + return f"Error generating final summary: {str(e)}" |
| 135 | + |
| 136 | +# Streamlit UI |
| 137 | +st.title("Decode Images and Videos with OCI GenAI") |
| 138 | +uploaded_file = st.sidebar.file_uploader("Upload an image or video", type=["png", "jpg", "jpeg", "mp4", "avi", "mov"]) |
| 139 | +user_prompt = st.text_input("Enter your prompt for analysis:", value="Describe the content of this image.") |
| 140 | +lang_type = st.sidebar.selectbox("Output Language", ["English", "French", "Arabic", "Spanish", "Italian", "German", "Portuguese", "Japanese", "Korean", "Chinese"]) |
| 141 | + |
| 142 | +if uploaded_file: |
| 143 | + if uploaded_file.name.split('.')[-1].lower() in ["png", "jpg", "jpeg"]: |
| 144 | + # Image Analysis |
| 145 | + temp_image_path = "temp_uploaded_image.jpg" |
| 146 | + with open(temp_image_path, "wb") as f: |
| 147 | + f.write(uploaded_file.getbuffer()) |
| 148 | + |
| 149 | + st.image(temp_image_path, caption="Uploaded Image", width=500) |
| 150 | + |
| 151 | + if st.button("Generate image Summary"): |
| 152 | + with st.spinner("Analyzing the image..."): |
| 153 | + try: |
| 154 | + encoded_image = encode_image(temp_image_path) |
| 155 | + llm_request = get_chat_request(encoded_image, user_prompt) |
| 156 | + llm_payload = get_chat_detail(llm_request,visionModel) |
| 157 | + llm_response = llm_client.chat(llm_payload) |
| 158 | + llm_text = llm_response.data.chat_response.choices[0].message.content[0].text |
| 159 | + st.success("OCI gen AI Response:") |
| 160 | + st.write(llm_text) |
| 161 | + except Exception as e: |
| 162 | + st.error(f"An error occurred: {str(e)}") |
| 163 | + elif uploaded_file.name.split('.')[-1].lower() in ["mp4", "avi", "mov"]: |
| 164 | + |
| 165 | + # Video Analysis |
| 166 | + temp_video_path = "temp_uploaded_video.mp4" |
| 167 | + video_html = f""" |
| 168 | + <video width="600" height="300" controls> |
| 169 | + <source src="data:video/mp4;base64,{base64.b64encode(open(temp_video_path, 'rb').read()).decode()}" type="video/mp4"> |
| 170 | + Your browser does not support the video tag. |
| 171 | + </video> |
| 172 | + """ |
| 173 | + st.markdown(video_html, unsafe_allow_html=True) |
| 174 | + with open(temp_video_path, "wb") as f: |
| 175 | + f.write(uploaded_file.getbuffer()) |
| 176 | + |
| 177 | + # st.video(temp_video_path) |
| 178 | + st.write("Processing the video...") |
| 179 | + |
| 180 | + frame_interval = st.sidebar.slider("Frame extraction interval (seconds)", 1, 10, 1) |
| 181 | + frames = extract_frames(temp_video_path, interval=frame_interval) |
| 182 | + num_frames = len(frames) |
| 183 | + st.write(f"Total frames extracted: {num_frames}") |
| 184 | + |
| 185 | + frame_range = st.sidebar.slider("Select frame range for analysis", 0, num_frames - 1, (0, num_frames - 1)) |
| 186 | + |
| 187 | + if st.button("Generate Video Summary"): |
| 188 | + with st.spinner("Analyzing selected frames..."): |
| 189 | + try: |
| 190 | + selected_frames = frames[frame_range[0]:frame_range[1] + 1] |
| 191 | + waiting_message = st.empty() |
| 192 | + waiting_message.write(f"Selected {len(selected_frames)} frames for processing.") |
| 193 | + # st.write(f"Selected {len(selected_frames)} frames for processing.") |
| 194 | + frame_summaries = process_frames_parallel(llm_client, selected_frames, user_prompt) |
| 195 | + # st.write("Generating final video summary...") |
| 196 | + waiting_message.empty() |
| 197 | + waiting_message.write("Generating final video summary...") |
| 198 | + final_summary = generate_final_summary(llm_client, frame_summaries) |
| 199 | + waiting_message.empty() |
| 200 | + st.success("Video Summary:") |
| 201 | + st.write(final_summary) |
| 202 | + except Exception as e: |
| 203 | + st.error(f"An error occurred: {str(e)}") |
0 commit comments