CNTK Object Recognition Video Tagging Tool
This tool provides end to end support for generating datasets for and validating Object Recognition Models with CNTK.
It supports the following scenarios:
- Computer Assisted Tagging of Objects In Video Using Custom implementation of Camshift Algorithm
- Export Tags to CNTK format for training a CNTK object recognition model.
- Running and validating a trained CNTK object recognition model on new videos to generate stronger models. (Windows only for now, GPU recommended)
To Use:
- Download the release binary.
- Extract and run the app
- Load a video.
- Configure the tagging job specifying the following preferences
- Frame Extraction Rate (Number of frames to tag per a second)
- Tagging Region Type (Rectangle, Point, Square)
- Export Frames Until (How far to export)
- Labels (Labels for tagging)
- Tag the video frame by frame.
- Export Video to CNTK Format
- Train model using Object Detection using FastRCNN.
- Load new video apply model to new video, validate tags, re-export, retrain
- Repeat step #8 on new videos until model preformance is satisfactory.
Upcoming Features
- Image Directory Tagging Support
- Tagging Project Management
- UI enhancements