Codes

We evaluate different existing trackers in our dataset and analyse the results. We make the scripts available and describe how to use them. Cython version of the three registraion based trackers [IC, ESM, NNIC] can be downloaded [here]. Original source codes [L1- APG, IVT and TLD] from the author's website were used for the Online Learned Trackers. The metrics we use to test a tracker's performance are.

To use the scripts to evaluate your algorithm the tracked result (T) should be in the format shown below. Ground truth (GT) also follows the same format.

frame ulx uly urx ury lrx lry llx lly
frame00001.jpg 41.00 324.00 209.00 306.00 227.00 514.00 59.00 537.00
frame00002.jpg 41.00 324.01 209.00 306.01 227.02 514.00 58.99 537.01
frame00003.jpg 41.00 324.01 209.01 306.01 227.01 514.01 58.98 537.00
frame00004.jpg 41.00 324.01 208.99 306.02 227.01 514.00 59.01 537.00
frame00004.jpg ... ... ... ... ... ... ... ...

Overall Success

`Input:`	`Ground Truth (GT) path`
	`Tracked Result (T) path`
	`Threshold for successful tracking (TH)`

`Output:`	`Overall Success (fraction (float))`

Average Drift

`Input:`	`Ground Truth (GT) path`
	`Tracked Result (T) path`
	`Threshold for successful tracking (TH)`

`Output:`	`Average Drift`

Speed Sensitivity

`Input:`	`Ground Truth (GT) path`
	`Tracked result (T) path`
	`Threshold for successful tracking (TH)`

`Output:`	`Speed Sensitivity (graph) with interframe motion`