일 | 월 | 화 | 수 | 목 | 금 | 토 |
---|---|---|---|---|---|---|
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 |
- coding
- 큐넷
- pytorch
- 딥러닝
- 우분투
- 자격증
- 실기시험
- 코딩
- Python
- 2020정보처리기사
- 기사시험
- NCS
- 파이썬
- Anaconda
- ubuntu
- 기사 실기
- 파이토치
- 국가자격증
- Apple
- 정보처리
- 정보처리기사
- 실기
- qnet
- python3
- 정보처리기사 실기
- keyboards
- torch
- 로지텍
- DEEPLEARNING
- Logitech
- Today
- Total
dhwiii's notepad | 딥 러닝, 코덱 일기장
[Video Codec] AI, RA, LD모드에 대해서 본문
All Intra/Random access/Low delay - in terms of video encoding, these terms define the GOP structure of encoded streams. Actually, they define the presence and frequency of key frames (I-frames) in the elementary video stream.
All intra
AI mode assumes that all frames in the elementary stream are key frames. The possibility to decode this stream starting from any frame is a benefit of this mode. Low encoding efficiency on the same bitrates is a big lack. Usually, this mode is used for temporal storing and editing video streams before final transcoding.
📌 모든 프레임이 키 프레임인 엘리먼트 스트림으로 구성되어 있다고 가정. 이 모드의 장점은 스트림을 어느 프레임에서든 시작해서 디코드할 수 있다는 것이며, 동일한 비트레이트에서 인코딩 효율이 낮은 것은 큰 단점이라고 볼 수 있다. 보통 AI 모드는 최종 트랜스코딩 전에 비디오 스트림을 일시적으로 저장하고 편집하는 데 사용된다.
Random access
RA mode assumes that elementary streams contain some key frames which can be used for starting the decoding process. In common these streams contain B-frames also. This mode is used for most cases of final encoding.
📌 엘리먼트 스트림에 일부 키 프레임이 포함되어 있으며, 이 키 프레임을 사용하여 디코딩 프로세스를 시작할 수 있다고 가정. 일반적으로 이 스트림들은 B-프레임도 포함하며, 대부분의 최종 인코딩 상황에서 이 모드가 사용된다.
Low delay
LD mode assumes minimal latency between capturing an image and its display after encoding-broadcasting-decoding pipeline. It may be achieved by avoiding Intra and Bidirectional frames in GOP structure. Also, low delay encoding usually assumes only one reference.
📌 이미지 캡처와 인코딩-방송-디코딩 파이프라인을 통한 디스플레이 사이의 최소 지연을 가정. 이는 GOP 구조에서 Intra 프레임과 양방향 프레임을 피함으로써 달성될 수 있으며, LD 모드는 보통 하나의 참조만을 가정합니다.
다시 말해서, AI, RA, LD 모드는 각 유저에 대한 환경을 가정하여 인-디코딩을 실시하기 위해 존재한다. 모든 영상을 화면내 예측으로 부호화하는 AI(All Intra), 1초마 다 화면내 슬라이스를 넣고 계층적 B 구조를 사용하는 RA(Random Access), 미래 영상을 참조하지 않는 LD(Low-delay) 부호화 구조 [2] 를 가지고 있는 것이 특징이다.
Reference
[1] https://vicuesoft.com/glossary/term/random_access_low_delay_all_intra/
[2] 호요성, Tech & Trend, "UHD 고화질 영상 서비스와 HEVC 영상 압축 표준"