The latest articles on DEV Community by Origin Data Lab (@kimhoonhoeglitch). https://dev.to/kimhoonhoeglitch https://media2.dev.to/dynamic/image/width=90,height=90,fit=cover,gravity=auto,format=auto/https:%2F%2Fdev-to-uploads.s3.us-east-2.amazonaws.com%2Fuploads%2Fuser%2Fprofile_image%2F4019456%2Fbbc5f3ed-02c4-45be-af0f-68d4a65dc25b.jpg https://dev.to/kimhoonhoeglitch en Origin Data Lab Tue, 06 Oct 2026 03:11:32 +0000 https://dev.to/kimhoonhoeglitch/technical-field-note-01-egocentric-rgb-imu-capture-evaluation-4534 https://dev.to/kimhoonhoeglitch/technical-field-note-01-egocentric-rgb-imu-capture-evaluation-4534 <h1> Technical Field Note #01: Egocentric RGB + IMU Capture Evaluation </h1> <p>Origin Data Lab has released a public technical field note documenting a first-party egocentric RGB + IMU capture evaluation.</p> <p>The evaluation demonstrates how we validate real-world multimodal capture before scaling field data production for Physical AI and robotics applications.</p> <p><a href="https://media2.dev.to/dynamic/image/width=800%2Cheight=%2Cfit=scale-down%2Cgravity=auto%2Cformat=auto/https%3A%2F%2Fdev-to-uploads.s3.us-east-2.amazonaws.com%2Fuploads%2Farticles%2F6ry1x0nwjkzpurxpqvl1.jpg" class="article-body-image-wrapper"><img src="https://media2.dev.to/dynamic/image/width=800%2Cheight=%2Cfit=scale-down%2Cgravity=auto%2Cformat=auto/https%3A%2F%2Fdev-to-uploads.s3.us-east-2.amazonaws.com%2Fuploads%2Farticles%2F6ry1x0nwjkzpurxpqvl1.jpg" alt=" " width="800" height="450"></a></p> <h2> What We Evaluated </h2> <ul> <li>1920×1080 egocentric RGB video at 30 fps</li> <li>Native and application-level IMU streams</li> <li>Measured sensor characteristics</li> <li>Video–IMU temporal alignment</li> <li>Structured task and interaction metadata</li> <li>Data integrity and reproducibility checks</li> </ul> <h2> Why This Matters </h2> <p>Real-world robotics data is not only about recording video.</p> <p>For downstream Physical AI and robotics applications, capture quality, sensor characteristics, synchronization, metadata, and reproducibility all affect whether a dataset can actually be used for training or evaluation.</p> <p>Our production workflow therefore validates the capture pipeline before scaling field collection.</p> <h2> Public Technical Evidence </h2> <p>The complete technical field note, including the evaluation details and technical evidence, is available on the Origin Data Lab website.</p> <p><a href="https://origindatalab.io/technical-field-notes/egocentric-rgb-imu-task-sample.html" rel="noopener noreferrer">https://origindatalab.io/technical-field-notes/egocentric-rgb-imu-task-sample.html</a></p> <h2> About Origin Data Lab </h2> <p>Origin Data Lab produces custom real-world multimodal datasets for Physical AI, robotics, embodied AI, and computer vision.</p> <p>We design and operate field data production around real tasks, real workers, and real operational environments.</p> <p>Our capabilities include egocentric video, RGB + IMU, stereo and multi-view capture, audio, timestamps, device metadata, metadata engineering, quality control, technical validation, recapture, and structured dataset delivery.</p> <p>We support project-based field production across Asia and Africa, with operations that can be configured around customer requirements.</p> <p>Website: <a href="https://origindatalab.io" rel="noopener noreferrer">https://origindatalab.io</a></p> ai robotics computervision machinelearning Origin Data Lab Tue, 07 Jul 2026 12:05:48 +0000 https://dev.to/kimhoonhoeglitch/how-we-built-a-real-world-bangladesh-rickshaw-traffic-dataset-for-physical-ai-1lgo https://dev.to/kimhoonhoeglitch/how-we-built-a-real-world-bangladesh-rickshaw-traffic-dataset-for-physical-ai-1lgo <p>Open datasets are essential for advancing AI, but truly diverse real-world traffic data remains difficult to find.</p> <p>To help address this gap, we built a Bangladesh Rickshaw Traffic Dataset designed for Physical AI, Computer Vision, Robotics, and Autonomous Driving research.</p> <p>The dataset captures complex urban traffic scenes where rickshaws, motorcycles, buses, trucks, pedestrians, and private vehicles interact naturally in challenging environments.</p> <h2> Why Bangladesh? </h2> <p>Most publicly available traffic datasets are collected in structured road environments where traffic behavior is relatively predictable.</p> <p>Bangladesh presents a very different scenario.</p> <p>Urban roads are shared by rickshaws, motorcycles, buses, trucks, private vehicles, street vendors, bicycles, and pedestrians. These complex interactions make the dataset particularly valuable for training AI systems that must operate in diverse real-world environments rather than controlled conditions.</p> <p>This diversity helps improve model robustness and generalization across different traffic scenarios.</p> <h2> Dataset Highlights </h2> <p>Some key characteristics of the dataset include:</p> <ul> <li>Real-world urban traffic captured in Dhaka</li> <li>Diverse traffic participants including rickshaws</li> <li>Human Ground Truth annotations</li> <li>Rich metadata for AI research</li> <li>Privacy-protected preview samples</li> <li>Commercial-ready organization</li> <li>Quality validation pipeline</li> <li>Dataset documentation</li> </ul> <h2> Metadata Design </h2> <p>We believe metadata is just as important as the video itself.</p> <p>Each clip is accompanied by structured metadata that helps researchers filter, search, and organize data efficiently.</p> <p>Examples include:</p> <ul> <li>Scene category</li> <li>Road context</li> <li>Object statistics</li> <li>Annotation status</li> <li>Quality indicators</li> <li>Privacy processing</li> <li>Capture information</li> <li>Dataset organization</li> </ul> <p>This reduces preprocessing time and makes the dataset easier to integrate into machine learning workflows.</p> <h2> Why We Opened a Free Sample </h2> <p>Many researchers and companies hesitate to evaluate datasets before purchasing them.</p> <p>To make evaluation easier, we released a free sample through multiple platforms so users can inspect the data quality before requesting larger commercial datasets.</p> <p>Our goal is transparency and long-term collaboration with the AI community.</p> <h2> Future Roadmap </h2> <p>This is only the beginning.</p> <p>Upcoming releases will include:</p> <ul> <li>Larger traffic datasets</li> <li>Night driving scenarios</li> <li>Rain and weather conditions</li> <li>Additional cities</li> <li>More Human Ground Truth annotations</li> <li>Enterprise-scale commercial datasets</li> <li>Expanded metadata for Physical AI and Robotics</li> </ul> <p>We plan to continuously improve both dataset quality and the supporting data pipeline.</p> <h2> Explore the Dataset </h2> <p>🌐 Website</p> <p><a href="https://origindatalab.io" rel="noopener noreferrer">Origin Data Lab</a></p> <p>💻 GitHub</p> <p><a href="https://github.com/kimhoonhoe-glitch" rel="noopener noreferrer">GitHub Repository</a></p> <p>🤗 Hugging Face</p> <p><a href="https://huggingface.co/origindatalab" rel="noopener noreferrer">Hugging Face Dataset</a></p> <p>📊 Kaggle</p> <p><a href="https://www.kaggle.com/origindatalab/datasets" rel="noopener noreferrer">Kaggle Datasets</a></p> ai computervision machinelearning opensource