i use custom HaarCasscade to train my model at this address "https://www.cs.auckland.ac.nz/~m.rezaei/Tutorials/Creating_a_Cascade_of_Haar-Like_Classifiers_Step_by_Step.pdf"
- Evaluation model (work on data set :(( )
- With IOU = 0.5, set model scaleFactors(multiScale) = 1.01 (1%)
ReCall: 65.7% , TP = 23, sum = 35
Predection: 95.83%, TP = 23 and FP = 1
- With IOU = 0.5, set mode; scaleFactors(multiScale) = 1.05 (5%)
ReCall: 40% , TP = 14, sum = 35
Predection: 100%, TP = 14 and FP = 0
- my model with 200 positive, 400 images negative (special condition)
- Evaluation model (work on data set :(( )
- With IOU = 0.5, set model scaleFactors(multiScale) = 1.01 (1%)
ReCall: 73.68% , TP = 154 and sum = 209
Predection: 96.85% , TP = 154 and FPS = 5
- With IOU = 0.5, set model scaleFactors(multiScale) = 1.05 (5%)
ReCall: 29.19% , TP = 61 and sum = 209
Predection: 95.31% , TP = 61 and FP = 3
- my model with 100 positive, 200 images negative (special condition)
- Evaluation model (work on data set :(( )
- With IOU = 0.5, set model scaleFactors(multiScale) = 1.01 (1%)
Recall: 50% , TP = 50 and sum = 100
Predection: 13.97% , TP = 50 and FP = 308
- With IOU = 0.5, set model scaleFactors(multiScale) = 1.05 (5%)
Recall: 51% , TP = 51 and sum = 100
Predection: 20.9% , TP = 51 and FP = 163
Note: Result so bad cause by the conditions of images
compare to result of haarcascade_russian_plate_number (on my data)
- With IOU = 0.5, set model scaleFactors(multiScale) = 1.01 (1%)
Recall: 43% , TP = 44 and sum = 108
Predection: 12.29% , TP = 44 and FP = 314 - With IOU = 0.5, set model scaleFactors(multiScale) = 1.05 (5%)
Recall: 35% , TP = 38 and sum = 108
Predection: 17.35% , TP = 38 and FP = 181
