Activity Forecasting:

Videos and presentation material


Activity Forecasting.
Kris M. Kitani, Brian D. Ziebart, Drew Bagnell and Martial Hebert.

European Conference on Computer Vision (ECCV 2012).

[Best Paper Runner-Up]

ECCV presentation video:

ECCV paper:

[PDF] (1.9 MB)

ECCV presentation material:
[] (54 MB - original, Mac Keynote)

[] (28MB - converted pptx)
PDF] (19.6 MB - no videos)



Destination Forecasting

Knowledge Transfer - Garden

Knowledge Transfer - VIRAT (Parking Lot)

Knowledge Transfer - VIRAT (UT)

Knowledge Transfer - VIRAT (Street)

30 second video spotlight

Supplementary Video

MORE DATA (2 scenes, ‘walk through scene’):


    Observed tracker trajectories and ground truth trajectories []


    Rectified (top-down) images []


    Feature maps []


    Learned reward weights []

    Precomputed reward function for forecasting []

    Precomputed empirical feature counts []

    List of base names []

    General explanation of data content [README.txt]

Activity Forecasting Demo Code:

There are two parts (IOC and OC) to the demo code, which can be used to generate the forecasting distribution over future paths. You can use the IOC code to learn an optimal set of reward parameters. The OC demo allows you to run just the test time inference with a set of reward parameters. Here are images generated by the code. You need to have OpenCV 2.3+ installed to run the programs.

Bird’s Eye View Image

Reward Function

Soft Value Function

Forecasting Distribution


Optimal Control (OC) Demo:

Main function [main.cpp]

Optimal control class header [oc.hpp]

Optimal control class methods [oc.cpp]

Input files (image, feature maps, reward weights and terminal points) []

Inverse Optimal Control (IOC) Demo:

Main function [main.cpp]

Inverse optimal control class header [ioc.hpp]

Inverse optimal control class methods [ioc.cpp]

Input files (basenames, images, features, trajectories, sample output) []