Fig. 1. The architecture of the SODMNet

Fig. 2. ResNet50 backbone network

Fig. 3. Bottleneck structure

Fig. 4. Comparison of depthwise convolution and pointwise convolution with traditional convolution

Fig. 5. Comparison of pointwise group convolution with traditional convolution

Fig. 6. MBConv structure

Fig. 7. Structure of FPN

Fig. 8. Annotated infrared stereo images

Fig. 9. Top-down view of object imaging with a parallel stereo camera setup

Fig. 10. Euclidean distance calculation for feature map descriptors

Fig. 11. Euclidean distance calculation between right and left feature map descriptors for corresponding and adjacent rows

Fig. 12. Intersection and union of the object bounding boxes

Fig. 13. Distance from predicted center to object bounding box

Fig. 14. Statistical results for dataset objects：（a） Initial dataset；（b） Supplement dataset

Fig. 15. Confusion matrix

Fig. 16. The prediction results of SODMNet in different scenarios