Video Predictions Visualizer
A relatively bare-bones interface for visualizing per-frame predictions and groundtruth labels on video data.
Click to play
Since everyone stores their predictions and groundtruth in a variety of formats, you'll need to write some code to allow the visualizer to access your data.
The basic steps are:
- Implement a DataLoader which implements the abstract methods in
data_loaders/data_loader.py. Here's an example that I wrote and is used for the demo above.
- "Register" your DataLoader
data_loaders/__init__.py, similar to how the
- Create a
.cfgfile for your data loader, similar to
video_visualizer.cfg. Note that the
DATA_LOADER_CONFIGconfig option will be passed as keyword arguments to your
Finally, you can run the server!
export FLASK_APP=video_visualizer.py FLASK_CONFIG=video_visualizer.cfg flask run