Multi-object tracking based on improved Fairmot framework

Table 1. Experimental hyperparameter settings
View table
View in Article
Table 1. Experimental hyperparameter settings
实验超参数未改进ReID Loss的实验改进ReID Loss的实验
优化器 Adam Adam
动量参数 0.9 0.9
Batchsize 6 6
初始学习率 1e-4 1e-4
训练轮数 45 70

Table 2. Multi-target tracking evaluation index

View table

View in Article

Table 2. Multi-target tracking evaluation index

指标	含义
MOTA	跟踪精度，衡量目标轨迹的保持程度
MOTP	衡量检测器的定位性能
IDF1	衡量跟踪器身份维持能力
MT	整个视频中超过80%的时间被正确跟踪的轨迹个数
ML	整个视频中不超过20%时间被正确跟踪的轨迹个数
IDS	目标身份的切换次数
FM	轨迹碎片的个数

Table 3. Ablation experiments of the three models tested on the MOT15 training set
View table
View in Article
Table 3. Ablation experiments of the three models tested on the MOT15 training set
模型 MOTA↑/% IDF1↑/% MOTP↑/% MT↑ ML↓ IDS↓ FM↓ Recall↑/% Prec↑/%
Fairmot 59.6 67.7 77.1 171 172 255 1 439 70.7 70.7
Fairmot+A 60.7 66.5 76.8 173 177 209 1 436 71 87.2
Fairmot+A+CL 62.9 64.3 77.5 255 123 232 1 173 78 84.3

Table 4. Comparative experiments of five models on the MOT15 test set
View table
View in Article
Table 4. Comparative experiments of five models on the MOT15 test set
模型 MOTA↑/% IDF1↑/% MT↑ ML↓ IDS↓
DMT^［27］ 44.5 49.2 250 159 684
MDP-Subcnn^［28］ 47.5 55.7 216 134 628
TRID^［29］ 55.7 61 293 105 351
Fairmot 60.6 64.7 343 79 591
Fairmot+A+CL 62.0 65.1 358 88 523

Tools

Get Citation

Copy Citation Text

Yi-fan XI, Li-ming HE, Yue LYU. Multi-object tracking based on improved Fairmot framework[J]. Chinese Journal of Liquid Crystals and Displays, 2022, 37(6): 777

Download Citation

EndNote(RIS)BibTex Plain Text

Set citation alerts for article

Save article for my favorites