Work

2024Vision

Pothole Detection (arXiv)

Co-author

I co-authored “Improved Pothole Detection Using YOLOv7 and ESRGAN” (arXiv:2401.08588), submitted in November 2023. Table 3 reports 0.947 precision and 0.826 recall for YOLOv7-tiny with ESRGAN on PNW dashcam frames.

Four PNW dashcam frames with red boxes drawn on labelled potholes, from the paper’s Figure 1.
Fig. 1Sample dashcam frames from the PNW set, with the labelled potholes marked. Figure 1 of Rout et al., arXiv:2401.08588. The boxes are the dataset labels, not model output.
ESRGAN upscales a low-resolution pothole crop, then YOLOv7 detects it against ground truth.
Fig. 2The paper’s pipeline: ESRGAN super-resolution, then YOLOv7 detection. Figure 2 of Rout et al., arXiv:2401.08588.
Three dashcam scenes, each shown low-resolution on the left and ESRGAN-upscaled on the right, with detector boxes.
Fig. 3Detector output on low-resolution frames (left) and the same frames after ESRGAN (right). Figure 4 of Rout et al., arXiv:2401.08588. Frames from the PNW dashcam set.
Three dashcam frames, each shown at low resolution and then after ESRGAN in the same place, with the detector’s boxes changing.
Fig. 4The same detector output, each frame alternated in place: low resolution, then after ESRGAN. Frames and boxes from Figure 4 of Rout et al., arXiv:2401.08588, alternated rather than set side by side.
Stack
YOLOv7, ESRGAN, Computer Vision, Published
Links
Paper, Code

Also