Crowd Counting Evaluation
- We focus on crowd counting on drone-captured scenes. In particular, given an input frame in bird-view, we require a counting method to estimate the number of people and density map in an image accurately.
- For this task, the performance is evaluated by both MAE and MSE scores, same as in [1]. Notably, the MAE score is used as the primary metric for ranking methods. The metrics are described in the following table.
MEASURE | PERFECT | DESCRIPTION |
MAE | 0 | Mean absolute error between the predicted number of people and ground-truth in evaluation |
MSE | 0 | Mean squared error between the predicted number of people and ground-truth in evaluation |
The evaluation code for crowd counting is available on the VisDrone github.
References:
[1] Wen, L., Du, D., Zhu, P., Hu, Q., Wang, Q., Bo, L., & Lyu, S. (2019). Drone-based Joint Density Map Estimation, Localization and Tracking with Space-Time Multi-Scale Attention Network.arXiv preprint arXiv:1912.01811.