Okoshi Takumi
Computer Vision Engineer at Orange
Yokohama, Japan
Summary
👤
Senior🎓
Top SchoolOkoshi Takumi is a Computer Vision Engineer and Kaggle GrandMaster with nine years of experience building production-ready vision and generative models from research to deployment. Based in Yokohama, he has driven model releases and advanced configurations for prominent open-source projects like MMDetection and Hugging Face Diffusers—contributing DINO and RTMDet variants, ControlNet integrations, and LoRA/IP-Adapter support. His industry work spans startups and larger teams (Orange, Rist, DeNA) focused on manga-specialized deep learning and applied NLP/vision, demonstrating an ability to ship domain-specific systems. Comfortable across Python and modern CV frameworks, he blends competition-grade research rigor with practical engineering to close the gap between SOTA models and real-world products. An often-overlooked strength is his track record of stabilizing complex toolchains and analysis utilities, helping teams reproduce and iterate on experiments more reliably.
9 years of coding experience
6 years of employment as a software developer
学士, Information Science/Studies, 学士, Information Science/Studies at 名古屋大学
Github Skills (22)
pytorch10
system-configuration10
python10
net10
diffusion-models10
stable-diffusion10
diffusers10
diffusion-probabilistic-models10
diffusion-probabilistic10
custom-configuration10
deeplearning-ai10
deep-learning10
image-generation10
lora10
object-detection10
Programming languages (2)
Github contributions (5)
open-mmlab/mmdetection
Nov 2019 - Jan 2024
OpenMMLab Detection Toolbox and Benchmark
Role in this project:
ML Engineer
ML EngineerContributions:8 reviews, 8 PRs, 23 comments in 4 years 2 months
Contributions summary:Okoshi primarily contributed to the release of new models and configurations within the MMDetection framework. They released DINO models with varying training schedules and backbone architectures like Swin-L. They also released an RTMDet-X p6 configuration and added new RTMDet models based on Swin and ConvNeXt backbones. Additionally, the user fixed a bug in an analysis tool.
retinanetbenchmarkfast-rcnnopenmmlabsemisupervised-learning
huggingface/diffusers
Jun 2022 - Feb 2024
🤗 Diffusers: State-of-the-art diffusion models for image, video, and audio generation in PyTorch and FLAX.
Role in this project:ML Engineer
ML EngineerContributions:10 reviews, 14 PRs, 67 comments in 1 year 8 months
Contributions summary:Okoshi primarily contributed to the development and enhancement of diffusion models within the Hugging Face Diffusers library. Their work involved implementing and refining features related to reference-based image generation and control mechanisms, as evidenced by commits focused on reference control and ControlNet integration. They also added and improved LoRA support for training and fine-tuning models, and integrated support for IP-Adapter Plus, expanding the library's capabilities in image generation.
pytorchartdeep-learningimage2imagestate-of-the-art
Find and Hire Top DevelopersWe’ve analyzed the programming source code of over 60 million software developers on GitHub and scored them by 50,000 skills. Sign-up on Prog,AI to search for software developers.