An unexpected journey into computer vision, or There and Back Again
summary
A team built a real-time map of road events from dashcam video by developing and deploying more efficient video object-detection models. The talk covers fundamentals of computer vision, available frameworks for training detection models, and the team’s approach using YOLO models and TensorFlow on Android, plus practical trials, difficulties, and surprising runtime bugs encountered.
AI-written summary · about
chapters (15)
- 0:00Introduction
- 4:29Talk goals and structure
- 5:11Dashcam project overview
- 8:46Existing model issues
- 9:10Deep learning frameworks comparison
- 16:05TensorFlow choice
- 17:08Neural network theory
- 21:46Building CNN model
- 27:07Model accuracy problems
- 28:13YOLO for object detection
- 31:31Mobile platform tools
- 32:12GPU and NNAPI usage
- 36:11Data preparation
- 39:51Detection and post-processing
- 44:27Conclusions and recommendations
related talks
24mBeyond the Frame: Using Computer Vision to Unlock the Potential of Video Data
46mML Use Cases at Think Big Analytics
01h 14mFoundations of Object Detection from an Expert's Overview
22mOpenCV and TensorFlow in practice
01h 29mComputer Vision: An (Un?)Expected Journey, with Keras and Tensorflow
26mSelf Driving cars with Opencv
44mCan AI See You? Your Introduction to Computer Vision and GenAI
33mComputer Vision, Not Just for Breakfast Anymore