7 papers
Structuring Open-Ended NAS: Semi-Automated Design Knowledge Structuring with LLMs for Efficient Neural Architecture Search
Yuiko Sakuma, Masakazu Yoshimura, Marcel Gröpl +4
Current neural architecture search (NAS) methods are often limited by their predefined, restrictive search spaces. While recent large language model (LLM)-assisted NAS methods enab…
LLM as a Tool, Not an Agent: Code-Mined Tree Transformations for Neural Architecture Search
Masakazu Yoshimura, Zitang Sun, Yuiko Sakuma +3
Neural Architecture Search (NAS) aims to automatically discover high-performing deep neural network (DNN) architectures. However, conventional algorithm-driven NAS relies on carefu…
SF-Mamba: Rethinking State Space Model for Vision
Masakazu Yoshimura, Teruaki Hayashi, Yuki Hoshino +2
The realm of Mamba for vision has been advanced in recent years to strike for the alternatives of Vision Transformers (ViTs) that suffer from the quadratic complexity. While the re…
Online Data Curation for Object Detection via Marginal Contributions to Dataset-level Average Precision
Zitang Sun, Masakazu Yoshimura, Junji Otsuka +2
High-quality data has become a primary driver of progress under scale laws, with curated datasets often outperforming much larger unfiltered ones at lower cost. Online data curatio…
Beyond RGB: Adaptive Parallel Processing for RAW Object Detection
Shani Gamrian, Hila Barel, Feiran Li +2
Object detection models are typically applied to standard RGB images processed through Image Signal Processing (ISP) pipelines, which are designed to enhance sensor-captured RAW im…
SemiISP/SemiIE: Semi-Supervised Image Signal Processor and Image Enhancement Leveraging One-to-Many Mapping sRGB-to-RAW
Masakazu Yoshimura, Junji Otsuka, Radu Berdan +1
DNN-based methods have been successful in Image Signal Processor (ISP) and image enhancement (IE) tasks. However, the cost of creating training data for these tasks is considerably…