PYIMAGECONF 2018

If you missed the Opening Reception you can register on site beginning at 8:00AM.

Image detection technology has come a long way in the last few years. The latest image segmentation models like Mask R-CNN not only detect the locations of objects in images, but they can draw outlines around objects and even separate out overlapping instances of objects. In this talk we'll start by learning about image segmentation and how the Mask R-CNN model is architected. Then we'll look at a project that counts lines of people in a live video stream using image segmentation. Next, we'll check out how to train a custom image segmentation model. Finally, we'll talk about the impact that image segmentation is having on the mapping industry and how it's helping nonprofit agencies in Africa reach people more effectively.

In this talk, we will provide a glimpse into the variety of real world applications in CVML that we (Big Vision LLC) have solved for our clients. Computer Vision is much more than importing your favorite library and training a model. Real world problems are messy with noisy inadequate data. Usually a combination of techniques is required to solve a problem. Sometimes you can generate synthetic data to solve a challenging problem. At other times a solution involves physics, geometry or even additional hardware.

In this talk I will survey a range of object detection algorithms, discussing their strengths, weaknesses, and the practical details you need to know to get them working in real applications. The talk will begin with an introduction to the very simple but popular HOG+SVM object detection algorithm. Many of the basic features of this algorithm appear in more advanced techniques, so having familiarity with it makes understanding the design decisions in more complex algorithms much easier to understand. From there we will survey other algorithms such as CNNs, regression trees, object saliency methods, and the Hough transform. I will discuss what these things do, why you should care, and how you can use them in your processing pipelines. Finally, practical advice for training detectors, how to construct good training datasets, and what to do when it's not working will be discussed. This talk will also serve as the introduction to the following day's workshop which will go into these topics in greater depth as we work through example applications of these techniques in Python.

Lunch will be served.

Computer vision can help manage a farm. This talk describes a computer vision pipeline distributed across 15 Raspberry Pi and Mac computers. A distributed pipeline allows each computer to do what it does most effectively. A Raspberry Pi can take pictures with the PiCamera and adjust camera settings, control supplemental lighting, crop, flip, grayscale and threshold images, as well as detect motion. A Mac can store and index images from many Raspberry Pi computers simultaneously and can perform more complex image processing like reading the changing digits of the water meter or using image classification to label a coyote or a raccoon in an image stream. This talk will share project design details and lessons learned along the way, as well as provide fun tips about PiCameras and sensor hardware.

Join us at the Hyatt Regency for drinks, networking, and socialization.

Lunch will be served.

Join us at the Hyatt Regency for drinks, networking, and socialization as we wrap up the very first PyImageConf.