Titan / utils /ingest_video.py
NEXAS's picture
Update utils/ingest_video.py
2ab65e0 verified
raw
history blame
4.63 kB
import os
import cv2
import chromadb
from chromadb.utils.embedding_functions import OpenCLIPEmbeddingFunction
from chromadb.utils.data_loaders import ImageLoader
# Initialize ChromaDB client and collection
path = "mm_vdb2"
client = chromadb.PersistentClient(path=path)
image_loader = ImageLoader()
CLIP = OpenCLIPEmbeddingFunction()
video_collection = client.get_or_create_collection(
name='video_collection',
embedding_function=CLIP,
data_loader=image_loader
)
def extract_frames(video_folder, output_folder):
"""
Extracts frames from all videos in the video_folder and saves them in the output_folder.
Args:
video_folder (str): Path to the folder containing video files.
output_folder (str): Path to the folder where extracted frames will be saved.
"""
if not os.path.exists(output_folder):
os.makedirs(output_folder)
for video_filename in os.listdir(video_folder):
if video_filename.endswith('.mp4'):
video_path = os.path.join(video_folder, video_filename)
video_capture = cv2.VideoCapture(video_path)
fps = video_capture.get(cv2.CAP_PROP_FPS)
frame_count = int(video_capture.get(cv2.CAP_PROP_FRAME_COUNT))
duration = frame_count / fps
output_subfolder = os.path.join(output_folder, os.path.splitext(video_filename)[0])
if not os.path.exists(output_subfolder):
os.makedirs(output_subfolder)
success, image = video_capture.read()
frame_number = 0
while success:
# Save frames at 0 seconds, every 5 seconds, and the last frame
if frame_number == 0 or frame_number % int(fps * 5) == 0 or frame_number == frame_count - 1:
frame_time = frame_number / fps
output_frame_filename = os.path.join(output_subfolder, f'frame_{int(frame_time)}.jpg')
cv2.imwrite(output_frame_filename, image)
success, image = video_capture.read()
frame_number += 1
video_capture.release()
def add_frames_to_chromadb(video_dir, frames_dir):
"""
Adds extracted frames from videos to the ChromaDB collection.
Args:
video_dir (str): Path to the folder containing video files.
frames_dir (str): Path to the folder containing the extracted frames.
"""
# Dictionary to hold video titles and their corresponding frames
video_frames = {}
# Process each video and associate its frames
for video_file in os.listdir(video_dir):
if video_file.endswith('.mp4'):
video_title = video_file[:-4]
frame_folder = os.path.join(frames_dir, video_title)
if os.path.exists(frame_folder):
# List all jpg files in the folder
video_frames[video_title] = [f for f in os.listdir(frame_folder) if f.endswith('.jpg')]
# Prepare ids, uris, and metadatas for ChromaDB
ids = []
uris = []
metadatas = []
for video_title, frames in video_frames.items():
video_path = os.path.join(video_dir, f"{video_title}.mp4")
for frame in frames:
frame_id = f"{frame[:-4]}_{video_title}"
frame_path = os.path.join(frames_dir, video_title, frame)
ids.append(frame_id)
uris.append(frame_path)
metadatas.append({'video_uri': video_path})
# Add frames to the ChromaDB collection
video_collection.add(ids=ids, uris=uris, metadatas=metadatas)
def initiate_video(video_folder_path):
"""
Initiates the video processing pipeline: extracts frames from videos
and adds them to the ChromaDB collection.
Args:
video_folder_path (str): Path to the folder containing video files.
Returns:
The ChromaDB collection with the added frames.
"""
try:
print("Starting video processing pipeline...")
# Define output folder for extracted frames
output_folder_path = os.path.join(video_folder_path, 'extracted_frames')
# Extract frames from videos
print("Extracting frames...")
extract_frames(video_folder_path, output_folder_path)
print("Frames extracted successfully.")
# Add frames to ChromaDB collection
print("Adding frames to ChromaDB...")
add_frames_to_chromadb(video_folder_path, output_folder_path)
print("Frames added to ChromaDB successfully.")
return video_collection
except Exception as e:
print(f"An error occurred during video processing: {e}")
return None