My Changes

This fork is primarily for viewing results from our network on the Charades dataset.

Note that it has been specifically designed to work on the videos resampled to 24 FPS, not the original videos.

Video Predictions Visualizer

A relatively bare-bones interface for visualizing per-frame predictions and groundtruth labels on video data.

Usage

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:

1. 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.
2. "Register" your DataLoader data_loaders/__init__.py, similar to how the MultiThumosDataLoader is, currently.
3. Create a .cfg file for your data loader, similar to video_visualizer.cfg. Note that the DATA_LOADER_CONFIG config option will be passed as keyword arguments to your DataLoader.

Finally, you can run the server!

export FLASK_APP=video_visualizer.py FLASK_CONFIG=video_visualizer.cfg
flask run