US Patent Application 18224501. VIDEO PROCESSING METHOD AND APPARATUS, DEVICE, AND STORAGE MEDIUM simplified abstract
Contents
VIDEO PROCESSING METHOD AND APPARATUS, DEVICE, AND STORAGE MEDIUM
Organization Name
TENCENT TECHNOLOGY (SHENZHEN) COMPANY LIMITED
Inventor(s)
Yuanyuan Zhao of Shenzhen (CN)
VIDEO PROCESSING METHOD AND APPARATUS, DEVICE, AND STORAGE MEDIUM - A simplified explanation of the abstract
This abstract first appeared for US patent application 18224501 titled 'VIDEO PROCESSING METHOD AND APPARATUS, DEVICE, AND STORAGE MEDIUM
Simplified Explanation
The patent application describes a video processing method performed by a computer device.
- The method involves inputting a target video and a video mask to an encoding model for feature extraction.
- This results in obtaining a first video feature vector and a second video feature vector.
- The index distribution of the first quantization feature vector in a discrete hidden space is determined.
- Based on the second video feature vector and the index distribution of the first quantization feature vector, a second quantization feature vector is determined in the discrete hidden space.
- The first quantization feature vector and the second quantization feature vector are then inputted to a decoding model.
- This results in obtaining a reconstructed video where the masked region of the target video is filled in accordance with the second quantization feature vector.
Original Abstract Submitted
This application discloses a video processing method performed by a computer device. The method includes: inputting a target video and a video mask to an encoding model for feature extraction to obtain a first video feature vector and a second video feature vector; determining an index distribution of the first quantization feature vector in a discrete hidden space composed on the basis of the first quantization feature vector; determining a second quantization feature vector in the discrete hidden space on the basis of the second video feature vector and the index distribution of the first quantization feature vector; and inputting the first quantization feature vector and the second quantization feature vector to a decoding model to obtain a reconstructed video, the reconstructed video referring to a video with a content of the masked region of the target video filled in accordance with the second quantization feature vector.