Task Feature Decoupling Model for Autonomous Driving Visual Joint Perception

Fig. 5. The visualized results of TFDJP,the multi-task model YOLOP,the single-task models PSPNet,YOLOv5, and SCNN on the BDD100K dataset. (a) The visualized results of YOLOP; (b) the visualized results of TFDJP; (c) the visualized results of PSPNet, YOLOv5, and SCNN

Download full size

Table 1. Ablation experiment results of the proposed module in object detection decoder of TFDJP model
View table
Table 1. Ablation experiment results of the proposed module in object detection decoder of TFDJP model
Method HSEM SIRM IoU perception prediction branch Recall /% mAP50 /%
A 89.2 76.5
B √ 90.3 78.4
C √ 90.6 78.3
D √ √ 91.1 78.8
E √ √ √ 91.7 79.2

Table 2. Ablation experiment results of different module in drivable area segmentation and lane detection decoder
View table
Table 2. Ablation experiment results of different module in drivable area segmentation and lane detection decoder
Method HRB LRB FFM
Edge
loss
mIoU /% Acc /% IoU /%
A 91.5 70.5 26.2
B √ 92.6 79.9 28.3
C √ 92.6 79.6 28.4
D √ √ √ 92.7 80.7 28.3
E √ √ √ √ 92.8 81.1 28.5

Table 3. Experiment results of object detection

View table

Table 3. Experiment results of object detection

Model	Recall /%	mAP50 /%	Frame rate /（frame/s）
Faster R-CNN^［24］	77.2	55.6	9.7
YOLOv5s^［5］	86.8	77.2	150.4
PP-YOLOE^［25］	82.3	59.9	47.0
MultiNet*^［15］	81.3	60.2	15.7
DLT-Net*^［16］	89.4	68.4	17.1
AMTNet*^［19］	88.3	77.8	68.8
YOLOP*^［17］	89.2	76.5	76.3
TFDJP*	91.7	79.2	67.5

Table 4. Experiment results of drivable area segmentation
View table
Table 4. Experiment results of drivable area segmentation
Model mIoU /% Frame rate /（frame/s）
PSPNet^［10］ 89.6 9.7
MultiNet*^［15］ 71.6 15.7
DLT-Net*^［16］ 71.3 17.1
AMTNet*^［19］ 88.3 68.8
YOLOP*^［17］ 91.5 76.3
TFDJP* 92.8 67.5

Table 5. Experiment results of lane detection
View table
Table 5. Experiment results of lane detection
Model Acc /% IoU /%
ENet^［26］ 34.12 14.64
SCNN^［11］ 35.79 15.84
ENet-SAD^27］ 36.56 16.02
AMTNet*^［19］ 73.60 26.91
YOLOP*^［17］ 70.50 26.20
TFDJP* 81.10 28.50

Tools

Get Citation

Copy Citation Text

Yue Wang, Jiale Cao. Task Feature Decoupling Model for Autonomous Driving Visual Joint Perception[J]. Laser & Optoelectronics Progress, 2024, 61(22): 2215007

Download Citation

EndNote(RIS)BibTex Plain Text

Set citation alerts for article

Save article for my favorites