Deep Learning in OpenCV
summary
OpenCV DNN implements forward-only inference for models from Caffe TensorFlow Torch and Darknet and runs on CPU OpenCL and the OpenVINO inference engine with a Vulkan (Vulkan/compute shader) backend in progress. It fuses conv+batchnorm+scale+ReLU and other patterns, reuses layer memory, and supports FP16 and OpenCL auto-tuning for tuned convolution kernels. A Tizen MobileNet‑SSD demo runs real‑time on Iris Pro GPU (speaker reports >60 FPS or ~30 FPS depending on backend), showing practical gains from these optimizations.
AI-written summary · about
chapters (10)
- 0:00Introduction
- 1:00What is OpenCV
- 2:34Deep Neural Networks
- 7:08Why
- 9:35Architecture
- 12:01Acceleration
- 14:25Network Optimization
- 14:53Layer Fusion
- 18:30OpenCL
- 20:49Vulcan backend
related talks
20mComputer Vision at the Edge with OpenVINO
57mDeep Learning with C++
25mOpenCL support by NNVM & TVM
55mA Tour of Deep Learning With C++
34mFaster deep learning solutions...
54mDeep Learning in JS
23mOptimize your network inference time with OpenVINO
39mRun Deep Learning models in the browser with JavaScript and ConvNetJS