Hello to everyone! I’ve problem about my custom traffic sign dataset i’ve already trained multiple times and those are my examples. My 300 epoch yolov5l trained dataset has not able to understand the no turn right. but another one which is 180 epoch trained the same way able to understand no right or no left. I’ve closed the mirror from hyp.startch so thats not the problem as u see. I’m sharing my dataset results + the test from gazebo. Can u help me about what’s my mistake ? . between of the dataset photos are not the same. But i believe the 300 epoch one is made by what yolov5 advice.
180 epoch sonkursunanar2 - Google Drive
300 epoch
sonkursun4 - Google Drive
180 epoch
300epoch
for 180 epoch from left to right
train/box_loss train/obj_loss train/cls_loss metrics/precision metrics/recall metrics/mAP_0.5 metrics/mAP_0.5:0.95 val/box_loss val/obj_loss val/cls_loss x/lr0 x/lr1 x/lr2
0.011771 0.0049416 0.0058296 0.96953 0.99129 0.99253 0.92036 0.0084473 0.0020838 0.0020874 0.004126 0.004126 0.004126
for 300 epoch from left to right
0.0082158 0.0021126 0.00030347 0.95946 0.97708 0.97673 0.87798 0.0086116 0.001293 0.00026458 0.00106 0.00106 0.00106
I can share the results zip from the drive or just can the share the drive link.
For 180 epoch best.pt results from gazebo (saga_donulmez = No turn right , sola_donulmez =no turn left
For 300 epoch
it does not read saga_donulmez from the away when its full sign only able to read that way.
