Hiroki Kobayashi
Computer Vision Researcher & AI Engineer

Hiroki Kobayashi

PhD in Engineering · Araya Inc.

I am a computer vision researcher and AI engineer specializing in industrial anomaly detection, representation learning, and generative modeling. My work focuses on building practical and robust machine learning methods for visual inspection, especially in manufacturing environments where real defective data are limited.

My research explores pseudo-defect generation, domain-specific pre-training, and feature modeling for anomaly detection, with the goal of improving both accuracy and generalization in real-world applications. I am particularly interested in bridging the gap between academic research and deployable AI systems.

In addition to research, I actively develop deep learning implementations in both Python and C++, including large-scale open-source projects based on PyTorch and LibTorch.

Anomaly Detection Visual Inspection Computer Vision Deep Learning Generative Modeling Representation Learning
2024
Anomaly Detection Based on Semi-Formula Driven Pre-training Dataset to Represent Subtle Difference and Anomaly Score First Author Acceptance Rate: 25.78%
Hiroki Kobayashi, Naoki Murakami, Naoto Hiramatsu, Takahiro Suzuki, Manabu Hashimoto
BMVC2024
2023
DRepT: Anomaly Detection Based on Transfer of Defect Representation with Transmittance Mask First Author Acceptance Rate: 54.76%
Hiroki Kobayashi, Manabu Hashimoto
IJCNN2023
pytorch_cpp
Open Source · C++ · LibTorch

PyTorch C++ Samples

A large-scale open-source project for deep learning implementations in C++ using LibTorch. Covers image classification, object detection, generative models, and a wide range of modern deep learning architectures.

View on GitHub →