Spaces:
Sleeping
Sleeping
File size: 12,649 Bytes
f25ff37 f279bf5 f25ff37 f279bf5 f25ff37 f279bf5 f25ff37 f279bf5 4a3a5b5 bdcf215 b5642bf f25ff37 bdcf215 f279bf5 bdcf215 f279bf5 bdcf215 f279bf5 b5642bf f279bf5 b5642bf f279bf5 b5642bf bdcf215 b5642bf bdcf215 b5642bf f279bf5 b5642bf f279bf5 b5642bf 6a8900c b5642bf f279bf5 b5642bf f25ff37 b5642bf bdcf215 b5642bf bdcf215 b5642bf bdcf215 b5642bf bdcf215 b5642bf bdcf215 b5642bf bdcf215 b5642bf bdcf215 b5642bf bdcf215 b5642bf bdcf215 f279bf5 bdcf215 4a3a5b5 bdcf215 4a3a5b5 bdcf215 4a3a5b5 bdcf215 f279bf5 b5642bf f279bf5 f25ff37 4a3a5b5 f25ff37 f279bf5 6a8900c 4a3a5b5 6a8900c b5642bf bdcf215 b5642bf bdcf215 b5642bf f279bf5 bdcf215 f279bf5 bdcf215 4a3a5b5 bdcf215 4a3a5b5 bdcf215 4a3a5b5 f279bf5 4a3a5b5 bdcf215 4a3a5b5 bdcf215 4a3a5b5 f279bf5 bdcf215 f25ff37 bdcf215 b5642bf f279bf5 b5642bf bdcf215 b5642bf 4a3a5b5 f279bf5 4a3a5b5 f279bf5 4a3a5b5 |
1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 31 32 33 34 35 36 37 38 39 40 41 42 43 44 45 46 47 48 49 50 51 52 53 54 55 56 57 58 59 60 61 62 63 64 65 66 67 68 69 70 71 72 73 74 75 76 77 78 79 80 81 82 83 84 85 86 87 88 89 90 91 92 93 94 95 96 97 98 99 100 101 102 103 104 105 106 107 108 109 110 111 112 113 114 115 116 117 118 119 120 121 122 123 124 125 126 127 128 129 130 131 132 133 134 135 136 137 138 139 140 141 142 143 144 145 146 147 148 149 150 151 152 153 154 155 156 157 158 159 160 161 162 163 164 165 166 167 168 169 170 171 172 173 174 175 176 177 178 179 180 181 182 183 184 185 186 187 188 189 190 191 192 193 194 195 196 197 198 199 200 201 202 203 204 205 206 207 208 209 210 211 212 213 214 215 216 217 218 219 220 221 222 223 224 225 226 227 228 229 230 231 232 233 234 235 236 237 238 239 240 241 242 243 244 245 246 247 248 249 250 251 252 253 254 255 256 257 258 259 260 261 262 263 264 265 266 267 268 269 270 271 272 273 274 275 276 277 278 279 |
import streamlit as st
import json
import faiss
import numpy as np
from sentence_transformers import SentenceTransformer
import base64
from PIL import Image
import io
import cv2
from insightface.app import FaceAnalysis
from moviepy.editor import VideoFileClip
from sklearn.cluster import DBSCAN
from sklearn.decomposition import PCA
import plotly.graph_objs as go
# Load models
@st.cache_resource
def load_models():
text_model = SentenceTransformer("all-MiniLM-L6-v2")
image_model = SentenceTransformer("clip-ViT-B-32")
face_app = FaceAnalysis(providers=['CPUExecutionProvider'])
face_app.prepare(ctx_id=0, det_size=(640, 640))
return text_model, image_model, face_app
text_model, image_model, face_app = load_models()
# Load data
@st.cache_data
def load_data(video_id, output_dir):
with open(f"{output_dir}/{video_id}_summary.json", "r") as f:
summary = json.load(f)
with open(f"{output_dir}/{video_id}_transcription.json", "r") as f:
transcription = json.load(f)
with open(f"{output_dir}/{video_id}_text_metadata.json", "r") as f:
text_metadata = json.load(f)
with open(f"{output_dir}/{video_id}_image_metadata.json", "r") as f:
image_metadata = json.load(f)
with open(f"{output_dir}/{video_id}_face_metadata.json", "r") as f:
face_metadata = json.load(f)
face_index = faiss.read_index(f"{output_dir}/{video_id}_face_index.faiss")
return summary, transcription, text_metadata, image_metadata, face_metadata, face_index
video_id = "IMFUOexuEXw"
output_dir = "video_analysis_output"
video_path = "avengers_interview.mp4"
summary, transcription, text_metadata, image_metadata, face_metadata, face_index = load_data(video_id, output_dir)
# Load FAISS indexes
@st.cache_resource
def load_indexes(video_id, output_dir):
text_index = faiss.read_index(f"{output_dir}/{video_id}_text_index.faiss")
image_index = faiss.read_index(f"{output_dir}/{video_id}_image_index.faiss")
return text_index, image_index
text_index, image_index = load_indexes(video_id, output_dir)
# Comprehensive face summarization
def create_comprehensive_face_summary(face_index, face_metadata, eps=0.5, min_samples=3):
face_embeddings = face_index.reconstruct_n(0, face_index.ntotal)
clustering = DBSCAN(eps=eps, min_samples=min_samples, metric='cosine').fit(face_embeddings)
face_clusters = {}
for i, label in enumerate(clustering.labels_):
if label not in face_clusters:
face_clusters[label] = []
face_clusters[label].append(i)
summary = []
for label, indices in face_clusters.items():
if label != -1: # Ignore noise points
cluster_appearances = [face_metadata[i] for i in indices]
cluster_summary = {
"cluster_id": f"cluster_{label}",
"face_count": len(indices),
"appearances": cluster_appearances,
"timeline": [
{"start": app['start'], "end": app['end']}
for app in cluster_appearances
],
"total_screen_time": sum(app['end'] - app['start'] for app in cluster_appearances),
"first_appearance": min(app['start'] for app in cluster_appearances),
"last_appearance": max(app['end'] for app in cluster_appearances)
}
summary.append(cluster_summary)
return summary, face_embeddings, clustering.labels_
# Create comprehensive face summary
face_summary, face_embeddings, face_labels = create_comprehensive_face_summary(face_index, face_metadata)
# Face cluster visualization
def plot_face_clusters_interactive(face_embeddings, labels, face_summary):
pca = PCA(n_components=3)
embeddings_3d = pca.fit_transform(face_embeddings)
unique_labels = set(labels)
colors = [f'rgb({int(r*255)},{int(g*255)},{int(b*255)})'
for r, g, b, _ in plt.cm.rainbow(np.linspace(0, 1, len(unique_labels)))]
traces = []
for label, color in zip(unique_labels, colors):
if label == -1:
continue # Skip noise points
cluster_points = embeddings_3d[labels == label]
cluster_info = next((c for c in face_summary if c['cluster_id'] == f'cluster_{label}'), None)
if cluster_info:
hover_text = [
f"Cluster {label}<br>"
f"Face count: {cluster_info['face_count']}<br>"
f"Total screen time: {cluster_info['total_screen_time']:.2f}s<br>"
f"First appearance: {cluster_info['first_appearance']:.2f}s<br>"
f"Last appearance: {cluster_info['last_appearance']:.2f}s"
for _ in cluster_points
]
else:
hover_text = [f"Cluster {label}" for _ in cluster_points]
trace = go.Scatter3d(
x=cluster_points[:, 0],
y=cluster_points[:, 1],
z=cluster_points[:, 2],
mode='markers',
name=f'Cluster {label}',
marker=dict(size=5, color=color, opacity=0.8),
text=hover_text,
hoverinfo='text'
)
traces.append(trace)
layout = go.Layout(
title='Face Clusters Visualization',
scene=dict(xaxis_title='PCA 1', yaxis_title='PCA 2', zaxis_title='PCA 3'),
margin=dict(r=0, b=0, l=0, t=40)
)
fig = go.Figure(data=traces, layout=layout)
return fig
# Search functions
def combined_search(query, text_index, image_index, text_metadata, image_metadata, text_model, image_model, n_results=5):
if isinstance(query, str):
text_vector = text_model.encode([query], convert_to_tensor=True).cpu().numpy()
image_vector = image_model.encode([query], convert_to_tensor=True).cpu().numpy()
else: # Assume it's an image
image_vector = image_model.encode(query, convert_to_tensor=True).cpu().numpy()
text_vector = image_vector # Use the same vector for text search in this case
text_D, text_I = text_index.search(text_vector, n_results)
image_D, image_I = image_index.search(image_vector, n_results)
text_results = [{'data': text_metadata[i], 'distance': d, 'type': 'text'} for i, d in zip(text_I[0], text_D[0])]
image_results = [{'data': image_metadata[i], 'distance': d, 'type': 'image'} for i, d in zip(image_I[0], image_D[0])]
combined_results = sorted(text_results + image_results, key=lambda x: x['distance'])
return combined_results[:n_results]
def face_search(face_embedding, face_index, face_metadata, n_results=5):
D, I = face_index.search(np.array([face_embedding]), n_results)
results = [face_metadata[i] for i in I[0]]
return results, D[0]
def detect_and_embed_face(image, face_app):
img_array = np.array(image)
faces = face_app.get(img_array)
if len(faces) == 0:
return None
largest_face = max(faces, key=lambda x: (x.bbox[2] - x.bbox[0]) * (x.bbox[3] - x.bbox[1]))
return largest_face.embedding
def create_video_clip(video_path, start_time, end_time, output_path):
with VideoFileClip(video_path) as video:
new_clip = video.subclip(start_time, end_time)
new_clip.write_videofile(output_path, codec="libx264", audio_codec="aac")
return output_path
# Streamlit UI
st.title("Video Analysis Dashboard")
# Sidebar with full video and scrollable transcript
st.sidebar.header("Full Video")
st.sidebar.video(video_path)
st.sidebar.header("Video Transcript")
transcript_text = transcription['transcription']
st.sidebar.text_area("Full Transcript", transcript_text, height=300)
# Main content
st.header("Video Summary")
# Face Clusters
st.subheader("Face Clusters")
for cluster in face_summary[:5]: # Display first 5 clusters
st.write(f"Cluster {cluster['cluster_id']}:")
st.write(f" Face count: {cluster['face_count']}")
st.write(f" Total screen time: {cluster['total_screen_time']:.2f} seconds")
st.write(f" First appearance: {cluster['first_appearance']:.2f} seconds")
st.write(f" Last appearance: {cluster['last_appearance']:.2f} seconds")
st.write(f" Timeline: {len(cluster['timeline'])} appearances")
st.write(" First 5 appearances:")
for app in cluster['timeline'][:5]:
st.write(f" {app['start']:.2f}s - {app['end']:.2f}s")
st.write("---")
# Face Cluster Visualization
st.subheader("Face Cluster Visualization")
fig = plot_face_clusters_interactive(face_embeddings, face_labels, face_summary)
st.plotly_chart(fig)
# Themes
st.subheader("Themes")
for theme in summary['themes']:
st.write(f"Theme ID: {theme['id']}, Keywords: {', '.join(theme['keywords'])}")
# Search functionality
st.header("Search")
search_type = st.selectbox("Select search type", ["Combined", "Face"])
if search_type == "Combined":
search_method = st.radio("Choose search method", ["Text", "Image"])
if search_method == "Text":
query = st.text_input("Enter your search query")
if st.button("Search"):
results = combined_search(query, text_index, image_index, text_metadata, image_metadata, text_model, image_model)
st.subheader("Search Results")
for result in results:
st.write(f"Type: {result['type']}, Time: {result['data']['start']:.2f}s - {result['data']['end']:.2f}s, Distance: {result['distance']:.4f}")
if 'text' in result['data']:
st.write(f"Text: {result['data']['text']}")
clip_path = create_video_clip(video_path, result['data']['start'], result['data']['end'], f"temp_clip_{result['data']['start']}.mp4")
st.video(clip_path)
st.write("---")
else:
uploaded_file = st.file_uploader("Choose an image...", type=["jpg", "jpeg", "png"])
if uploaded_file is not None:
image = Image.open(uploaded_file)
st.image(image, caption="Uploaded Image", use_column_width=True)
if st.button("Search"):
results = combined_search(image, text_index, image_index, text_metadata, image_metadata, text_model, image_model)
st.subheader("Image Search Results")
for result in results:
st.write(f"Type: {result['type']}, Time: {result['data']['start']:.2f}s - {result['data']['end']:.2f}s, Distance: {result['distance']:.4f}")
clip_path = create_video_clip(video_path, result['data']['start'], result['data']['end'], f"temp_clip_{result['data']['start']}.mp4")
st.video(clip_path)
st.write("---")
elif search_type == "Face":
face_search_type = st.radio("Choose face search method", ["Select from clusters", "Upload image"])
if face_search_type == "Select from clusters":
cluster_id = st.selectbox("Select a face cluster", [cluster['cluster_id'] for cluster in face_summary])
if st.button("Search"):
selected_cluster = next(cluster for cluster in face_summary if cluster['cluster_id'] == cluster_id)
st.subheader("Face Cluster Search Results")
for appearance in selected_cluster['appearances'][:5]: # Show first 5 appearances
st.write(f"Time: {appearance['start']:.2f}s - {appearance['end']:.2f}s")
clip_path = create_video_clip(video_path, appearance['start'], appearance['end'], f"temp_face_clip_{appearance['start']}.mp4")
st.video(clip_path)
st.write("---")
else:
uploaded_file = st.file_uploader("Choose a face image...", type=["jpg", "jpeg", "png"])
if uploaded_file is not None:
image = Image.open(uploaded_file)
st.image(image, caption="Uploaded Image", use_column_width=True)
if st.button("Search"):
face_embedding = detect_and_embed_face(image, face_app)
if face_embedding is not None:
face_results, face_distances = face_search(face_embedding, face_index, face_metadata)
st.subheader("Face Search Results")
for result, distance in zip(face_results, face_distances):
st.write(f"Time: {result['start']:.2f}s - {result['end']:.2f}s, Distance: {distance:.4f}")
clip_path = create_video_clip(video_path, result['start'], result['end'], f"temp_face_clip_{result['start']}.mp4")
st.video(clip_path)
st.write("---")
else:
st.error("No face detected in the uploaded image. Please try another image.") |