Searcharxiv⌕ Search

arXiv subjects

Youngjin Jo

Publications and source records attributed to Youngjin Jo.

2 recordsLinked to original sources

Benchmarking Vendor-Guided Optimizations on Edge NPUs: From NPU Only to End-to-End Performance. Extended technical report

Edge NPU performance depends not only on hardware capabilities, but also on model transformation and the overall deployment pipeline. We propose a benchmarking methodology that separates these effects by comparing three configurations: Baseline, Vendor Guided NPU Only (VG-NPU), and Vendor Guided End-to-End (VG-E2E), where VG-NPU preserves common model I/O and host-side processing while VG-E2E follows the vendor-native deployment pipeline. We apply this methodology to four edge NPUs and three vision workloads. VG-NPU improves both accuracy and Runtime-call latency in 10 of 12 NPU--workload combinations, whereas VG-E2E reduces System-level latency in all combinations but improves accuracy in only four. Moreover, NPU-level performance rankings do not always match End-to-End rankings, and VG-E2E provides an additional 10.4--70.7\% reduction in System-level latency over VG-NPU. These results demonstrate the importance of evaluating vendor-guided optimization separately at the controlled model-transformation and full deployment levels when comparing heterogeneous commercial Edge NPU platforms in practice.

cs.AI↗

TomoReal: Tomographic Displays

Since the history of display technologies began, people have dreamed an ultimate 3D display system. In order to get close to the dream, 3D displays should provide both of psychological and physiological cues for recognition of depth information. However, it is challenging to satisfy the essential features without sacrifice in conventional technical values including resolution, frame rate, and eye-box. Here, we present a new type of 3D displays: tomographic displays. We claim that tomographic displays may support extremely wide depth of field, quasi-continuous accommodation, omni-directional motion parallax, preserved resolution, full frame, and moderate field of view within enough eye-box. Tomographic displays consist of focus-tunable optics, 2D display panel, and fast spatially adjustable backlight. The synchronization of the focus-tunable optics and the backlight enables the 2D display panel to express the depth information. Tomographic displays have various applications including tabletop 3D displays, head-up displays, and near-eye stereoscopes. In this study, we implement a near-eye display named TomoReal, which is one of the most promising application of tomographic displays. We conclude with the detailed analysis and thorough discussion for tomographic displays, which would open a new research field.

cs.GR↗