You know I am lazy in learning new language, that's why I sticks to MATLAB and use the pretrained model to solve my easy problem.
Before using YOLO V3, I have downloaded YOLO V5, but I haven't find the the way to retrain in MATLAB. maybe later. After I use v3, I got the tricks.
1 Tricks
When you reuse the demo, it will pop up different errors. The reason is you need same basic function.
Remember to add in your current file.
generateYOLOv3Detections.m; generateTargets.m
2 Step to Retrain
DA
Loss curve..
use sequeese net+ yolo v3
Training Result
For rebar detection, we consider the IOU>0.5 and set it's precision to 1, therefore ,you should code your own precision based on your predition and groundtruth.
By YOLO V3. very good..
The cost time is 0.09s
1 Rebar feature
using "Automatic hyperbola detection and fitting in GPR B-scan image - ScienceDirect" method to find the hyperbolic curve.
Therefore, we need filter..
2 noise point
Wrong detection
Hey guys, you must know, we can detect the rebar and it's point by testing thounds of samples.
see the following.