May 2022 Summaries

Mark McQuade shares his experience of using computer vision to develop a model that detects and classifies different types of boxing punches, such as jabs, crosses, hooks, and uppercuts, during a workout session. Initially, the project involved capturing data from a 12-minute boxing session and annotating 239 images, which were then processed using the Roboflow platform to create a dataset. The model underwent several iterations, including the addition of a 'no punch' class and refinement of bounding boxes to improve accuracy. Despite initial challenges, the third version of the model performed better in distinguishing punches, although it still struggled with differentiating similar punch types like hooks and crosses. Further improvements could be made by increasing the dataset size and incorporating more pre-processing techniques. The project culminated in using Roboflow's video inference tool to analyze an entire boxing round, with eventual plans to calculate punch statistics to enhance workout effectiveness.

Roboflow, a remote-first company, prioritizes creating a magnetic environment for exceptional talent by hosting biannual on-site gatherings to foster team relationships and collaboration beyond virtual interactions. These on-sites, including a recent one in Miami, combine structured activities like Beach Olympics and computer vision projects with open, relaxed time to balance work and play, encouraging team members to connect with colleagues from different departments. By setting clear expectations and goals, offering travel stipends, and incorporating fun and meaningful activities, Roboflow enhances cross-functional trust and camaraderie among its team members. Additionally, the company emphasizes the importance of individuality and inclusion, encouraging diverse personalities to thrive during these gatherings. Feedback is actively sought post-onsite to continually improve the experience, with the Miami event rated highly by participants, highlighting Roboflow's commitment to nurturing a supportive and dynamic company culture.

Ultralytics released YOLOv8 Oriented Bounding Boxes in 2024, which outperforms YOLOv5 by utilizing rotated bounding boxes to more accurately detect angled objects, as illustrated with a pill detection dataset. The tutorial details the use of YOLOv5-obb for oriented object detection, starting with exporting an annotated dataset from Roboflow, a tool that allows for free dataset management and format conversion for up to 10,000 images. The process involves setting up the YOLOv5-OBB repository, installing necessary packages, and training the model with the dataset structured in a specific format, using code snippets provided in a Google Colab notebook. After training, detection is performed using the trained model, and users are encouraged to explore additional datasets available on Roboflow Universe for further practice.

Roboflow's April 2022 Changelog highlights several significant updates and expansions across its platform, focusing on enhancing user experience and capabilities in computer vision applications. The company introduced a new onboarding flow for easier team integration, launched instance segmentation as a new project type, and added segmentation as an export option. Offline support for enterprise GPU inference was also introduced, alongside private betas for native OAK, iOS, and React Native deployments. The Roboflow Universe saw the release of 65,000 datasets and a new search engine for improved user navigation. Company highlights include an onsite meeting in Miami, the addition of three team members, and multiple new job openings. The community contributed numerous innovative projects using Roboflow's resources, covering topics from intelligent invoice parsing to traffic management with UAVs.

The blog post by Mohamed Traore and Kelly M. discusses a project centered on creating a multi-label classification dataset to identify favorite edibles associated with Cinco de Mayo, such as tacos and beer, using computer vision techniques. The article clarifies that Cinco de Mayo commemorates Mexico's victory at the Battle of Puebla in 1862, not to be confused with Mexico's Independence Day. Multi-label classification is highlighted as a method that allows models to learn and predict multiple classes for a given image, which is advantageous when identifying complex settings or simply recognizing objects without focusing on their exact location. The authors detail the use of diverse datasets to ensure a robust model and describe the testing process using Roboflow's rfWidget and Hosted Inference API, which outputs JSON values indicating confidence scores for each predicted class. The flexibility of JSON format is emphasized for its compatibility with various programming languages and its utility in database interactions. The post concludes with a call to explore more projects on Roboflow Universe, encouraging responsible celebration of the holiday.

XR Tech, a UK-based provider of extended reality technologies, employs machine vision to enhance industrial safety and operational efficiency. Utilizing devices like RealWear and Hololens, XR Tech equips workers with body-mounted cameras that stream data to machine vision models trained on the Roboflow platform, forming a smart safety system that detects personal protective equipment (PPE) compliance and high-risk hazards. The system's rapid response capability improves safety in high-risk environments by addressing issues such as improper PPE usage. To develop effective models, XR Tech emphasizes good training practices, data hygiene, and targeted data collection, utilizing an open-source hard hat dataset to speed model development. Their mobile app processes live-streamed data, sending it to models hosted on Roboflow's endpoint, with the results used to inform safety measures. This integration of cutting-edge technology and best practices in machine vision is yielding significant advancements in industrial applications.