4.4.2. Influence of Parameters

To devise the temporally aligned representations, we equally divide each walking cycle into *N* segments according to the corresponding sinusoid curve. Here we evaluate the effects of the number of segments *N*, as reported in Table 3. The results show that STAR with *N* = 8 obtains the best performance, thus we divide a walking cycle into 8 segments for quantitative comparison in Section 4.2.

**Table 3.** Evaluation of the number of segments in a walking cycle on iLIDS-VID. "N" in "STAR\_N" is the number of segments. Bold values indicate the best performance.


We also evaluate the influence of the number of superpixels in Table 4, which is set as 100 in our previous experiments. Note that we perform superpixel segmentation in both motion information extraction and image representation. The results show that STAR with 100 superpixels obtains the best performance, thus we set it as 100 in our experiments.

**Table 4.** Evaluation of the number of superpixel on iLIDS-VID. Bold values indicate the best performance.

