Video Clip Extraction Using Frame Sampling and Pre-trained Scoring Models
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing video clip extraction methods require complete video processing, leading to high calculation workloads and long extraction durations, negatively impacting user experience.
Innovation Solution
A method that samples video frames and uses pre-trained frame feature extraction and scoring models to quickly determine scores, allowing for efficient extraction of target video clips without needing the entire video to be loaded.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If video clip extraction is performed by completely obtaining the video and dividing it into multiple video clips according to content, then accurate video clip extraction can be achieved, but large amount of calculation is required leading to long extraction duration
Solution Approach 1:
The patent applies preliminary action by pre-training frame feature extraction models and scoring models before actual video clip extraction. The models are trained offline to learn video content features and scoring patterns, so that during online extraction, only frame sampling and model inference are needed, dramatically reducing extraction duration while maintaining accuracy
Solution Approach 2:
The patent segments the video processing task into discrete frame-level operations. Instead of processing the entire video as one unit, it samples individual frames, extracts features for each frame independently, scores them separately, and then reconstructs video clips from scored frames. This segmentation enables parallel processing and reduces computational complexity
2Measurement precision
If video clip extraction is performed by completely obtaining the video, then complete video content analysis can be achieved, but large amount of calculation is required leading to high computational complexity
Solution Approach 1:
The patent applies partial action by sampling only N key frames from the video instead of processing every frame. By strategically selecting representative frames and using pre-trained models to score them, the system achieves sufficient video content analysis with significantly reduced computational workload compared to complete frame-by-frame processing
Solution Approach 2:
The patent uses copying by employing pre-trained models that have already learned video feature extraction and scoring patterns from extensive training data. These models are copied from offline training and applied online without retraining, allowing rapid inference on sampled frames while maintaining analysis completeness
Data Source
AI summary
A method for video clip extraction includes: obtaining a video, and sampling the video to obtain N video frames, wherein N is a positive integer; inputting the N video frames to a pre-trained frame feature extraction model to obtain a feature vector of each video frame in N video frames; determining scores of the N video frames based on a pre-trained scoring model; and extracting target video clips from the video based on the scores of the N video frames.