정보
활동
-
With DLSS 3.5 !!!! + Ray Reconstruction, Ray tracing, Reflex, and DLAA. 😘 이 소식을 전하고 싶어서 입이 얼마나 근질근질했는지 모릅니다. 많은 분들의 노력 끝에 퍼스트 디센던트에서 DLSS 3.5를…
With DLSS 3.5 !!!! + Ray Reconstruction, Ray tracing, Reflex, and DLAA. 😘 이 소식을 전하고 싶어서 입이 얼마나 근질근질했는지 모릅니다. 많은 분들의 노력 끝에 퍼스트 디센던트에서 DLSS 3.5를…
추천한 사람: Yongha Kim
-
TGS 용 트레일러가 공개되었습니다! https://lnkd.in/d2iG-JtB (영어 버전) https://lnkd.in/dNdjH2qd (일본어 버전) #NEXONGAMES #TheFirstDescendant #TGS #넥슨 #넥슨게임즈 #퍼스트디센던트
TGS 용 트레일러가 공개되었습니다! https://lnkd.in/d2iG-JtB (영어 버전) https://lnkd.in/dNdjH2qd (일본어 버전) #NEXONGAMES #TheFirstDescendant #TGS #넥슨 #넥슨게임즈 #퍼스트디센던트
추천한 사람: Yongha Kim
경력 및 학력
논문·저서
-
Generating globally unique identifiers for game objects (GPG6)
Charles River Media
Game Programming Gems series are well known guide books for advanced game programmers.
I wrote an article in Networking Section of Game Programming Gems 6.
-
A Fast Vector Quantization using Subregion - based Caches of Codeword Indexes
정보과학회논문지 : 소프트웨어 및 응용 제28권 제4호, 2001.4, 369-379 (11 pages)
A fast codebook generation method using the sub-region based caches of codeword indexes is proposed for vector quantization. The proposed method exploits the localization property that the proximate input vectors are usually represented by a specific part of codewords in the codebook. Initially, all training input vectors are partitined into multiple disjoint subregions of input vectors according to their proximities. A cache of the codeword indexes is assigned to each individual subregion…
A fast codebook generation method using the sub-region based caches of codeword indexes is proposed for vector quantization. The proposed method exploits the localization property that the proximate input vectors are usually represented by a specific part of codewords in the codebook. Initially, all training input vectors are partitined into multiple disjoint subregions of input vectors according to their proximities. A cache of the codeword indexes is assigned to each individual subregion, where the cache has a full size of codeword indexes in the very begining of learning and it maps the input vectors into the corresponding codewords through the codeword indexes. Due to the localization property, non-matched codeword indexes will be occurred and they are purged from the cache because they will not be refereneced any more. As the iteration goes on, a small number of codeword indexes that are referenced by the input vectors in the subregion are remained in the cache. The proposed scheme reduces the codebook generation time greatly because it computes the distortion with only relavant codeword indexes in the corresponding cache. It does not also degrade the recovered image quality too much because the purging policy is conservative and safe by discarding half of the non-matched codeword indexes at every iteration and the furthest codeword index from the prototype of the subregion first. Simulation results show that the proposed method speeds up the codebook generation time (or encoding time) by 2.6-5.4 times (or 3.7-18.8 times), respectively, than the LBG full search method, without sacrificing the recovered image quality.
다른 저자
프로젝트
외국어
-
Korean
원어민 수준
-
Japanese
상급
-
English
상급
다른 비슷한 프로필
대한민국 Yongha Kim님의 동명이인
-
Yongha Kim
Applied Scientist at KC ML2
-
Yongha Kim
Samsung Electronics Principal Engineer
-
Yongha Kim
충남대학교 Assistant Professor
-
Yongha Kim
Deloiite Anjin CPA
LinkedIn에 가입한 대한민국 내 Yongha Kim님의 동명이인 34명
Yongha Kim님의 동명이인