UEC Dataset (Choreographed Videos)

1. • Quad sequence [QUAD.MP4.zip] (254 MB)
   

2. • Park sequence [PARK.MP4.zip] (1.56 GB)
   

YouTube Dataset (Non-choreographed Videos)

1. •Mountain biking ( link )
   

2. •Snowboarding ( link )
   

3. •Slope style ( link )
   

4. •Horseback riding ( link )
   

5. •Skiing ( link )
   

6. •Longboard Surfing ( link )
   

(You might find this helpful...Chrome YouTube Downloader)

Ground Truth

Video segment labels (egoaction_gt.zip)

Code

1. •C/C++ Class for online inference with Dirichlet Process Mixture Models (onlineDPM-1.0.zip)
   

2. •C/C++ Class wrapper for the OpenCV sparse optical flow (optflow-1.0.zip)
   

3. •C/C++ Class for computing the motion histogram given a set of point correspondences (mohist-1.0.zip)
   

4. •C/C++ Sample code that stores motion features to a text file and prints the bag-of-features to the screen every 60 frames  (extractmohist-1.0.zip)
   

FAQs

Do results depend on my image size?

Yes, the flow magnitude and flow variance bin thresholds are optimized for WVGA (840 x 480)

Does optflow require any libraries?

Yes, you need to have OpenCV. The code should work with version 2.0 and above.

Do I need to calibrate for lens distortion?

No, parameters are actually optimized for the GoPro lens with radial distortion.

Fast Unsupervised Ego-Action Learning for First-Person Sports Videos.
Kris M. Kitani (UEC Tokyo), Takahiro Okabe (U.Tokyo), Yoichi Sato (U.Tokyo), Akihiro Sugimoto (NII). CVPR 2011. pp. 3241-3248.

[PAPER] (1.9 MB)
[POSTER] (2.4 MB)