Organizers:
Alexander Bovyrin
Nikita Manovich
Sergei Nosov
Dmitry Kurtaev
Time: 13:30-17:30 (Half Day — Afternoon)
Description:
Today’s Computer Vision algorithms are mostly powered with Deep Learning technique, which is both compute- and data-hungry. Modern hardware is usually heterogeneous and is becoming harder and harder to program efficiently. OpenCV 4.0 comes with new functionality to address these new challenges and provide developers with convenient APIs to handle the complexity. You may train the neural network model in one of popular deep learning frameworks (Caffe/TensorFlow/Darknet/Torch/ONNX format compatible frameworks) and run it using OpenCV without dependency on origin framework. The tutorial covers OpenCV 4.0 features introduction, deep learning module usage with code samples in C++, Python, Java and JavaScript (emscripten bindings). We will also make a review of different computation backends for deep networks such as OpenCL and Intel® Inference Engine. There will also be a practical hands-on session where participants will play with the new functionality. In particular participants will know:
- How to run deep networks on Android device with OpenCV 4.0;
- How to run deep networks in browser with OpenCV 4.0;
- Custom deep learning layers support in OpenCV 4.0.
Also, given that data is becoming critically important in this domain, OpenCV now hosts Computer Vision Annotation Tool (CVAT) which is web-based, free, online, interactive video and image annotation tool for computer vision. Easy deployment using docker, friendly user interface, optimized workflows to annotate data for typical computer vision tasks like object detection, image classification, semantic and instance segmentation make it popular among researchers around the world. There will be practical session on CVAT.
We will also provide update on Open Model Zoo (pre-trained deep learning models and samples) that can be downloaded from https://github.com/opencv/open_model_zoo and cover new features of Intel® Deep Learning Deployment Toolkit (CNN quantization tool, 3D conv, etc) with performance characteristics and practical samples.
We will also present our view of the future development of OpenCV and tools.
A G E N D A
| 13:40 – 15:30 | OpenCV 4.0 deep dive. features introduction, deep learning module usage with code samples in C++, Python, Java and JavaScript (emscripten bindings) . How to run deep networks on Android device with OpenCV 4.0; How to run deep networks in browser with OpenCV 4.0; Custom deep learning layers support in OpenCV 4.0. Make a review of different computation backends for deep networks such as OpenCL and Intel® Inference Engine. |
| 15:30 – 16:00 | Open Model Zoo. New models and performance. Trainable models. |
| 16:00 – 16:30 | Coffee break. |
| 16:30 – 17:00 | CNN compression. Int8 models and performance. Int1 models and performance. |
| 17:00 – 17:30 | Computer Vision Annotation Tool (https://github.com/opencv/cvat) |
Workshop Slides
CVPR 2019 – OpenCV workshop slides can be downloaded using the links given below.
