[Paper]

정민우 XDA: Representation-Aware Exponent-Direct Addressing for Efficient Shared Nonlinear Units석종민 FlowTT: Exploiting Computation Flow Reuse in Irregular Tensor-Train Embedding} 석종민 축하합니다!!!!

1저자 김권중 학생 뿐 아니라 공저자 이두열, 양재홍 학생도 수고 많았습니다. See Through the Occlusions: Few-Shot Gaussian Splatting with Layered Amodal Supervision

LC-Mamba: Local and Continuous Mamba with Shifted Windows for Frame Interpolation

Hybrid Embedding Framework for Memory-Efficient Recommendation Systems

Viewpoint-Adaptive Collage-based Streaming for 4K Light Field Video

Micro-Cells Enabling Spatio-Angular Joint Attention for High-Resolution Light Field Editing Overview of the proposed system. The input includes the pre-editing LF, and one edited SAI. Geometry information is inferred using mono depth estimation to generate a disparity map, for warping-based propagation. The edited LF is refined using a ViT-based composition network to address artifacts like …

2024.11 김권중 학생의 라이트필드 처리를 위한 트랜스포머 구조에 대한 논문 IEEE Access 에 게재되었습니다. 더 보기 »

Accelerating CNN Training with Concurrent Execution of GPU and Processing-In-Memory The baseline system architecture in this paper, consisting of a host and a PIM. The host is a GPU similar in design to NVIDIA’s V100, and the main memory is three-dimensional (3D) stacked DRAM (HBM). The PIM consists of a single-instruction-multiple- data (SIMD) operator and …

2024.10 CNN 학습을 위한 가속 방법에 관한 논문이 IEEE Access 에 게재되었습니다. 더 보기 »

The proposed 3D reconstruction system has two main stages: iterative depth predictions for occluded regions and 3D reconstruction using these predictions. Initially, conventional monocular depth estimation is applied, followed by iterative amodal mask generation (AMG) and amodal depth estimation (ADE) to generate amodal depths, enriching the 3D reconstruction with occlusion information.