File | Description | Shape |
---|---|---|
dataset_events_p.npy |
event polarity | (n, ) short |
dataset_events_t.npy |
event time in seconds | (n, ) float64 |
undistorted_events_xy.npy |
event coordinates. 1st row is width, 2nd row is height. | (n, 2) float32 |
The data are consistent as the description in README. Polarity is not used in our model, while we keep it in case users want to compare against other methods.
The precomputed inputs for MVSEC, DSEC, EVIMO can be downloaded from this drive. The data format is exactly the same as the demo data.
python main.py
It will generate flow predictions and uncertainties. It also generates a flow prediction video at output.mp4
. It is helpful to read the comments at main.py to understand the usage.