PROJECT: PERFORMANCE COMPARISON OF DEEP LEARNING MODELS IN VEHICLE DETECTION FROM INPUT IMAGE
Data source: Victoria University Staff Library
Dataset link: https://vustaff-my.sharepoint.com/:u:/g/personal/e5112162_vu_edu_au/Ef0WSbY7uQtPi1HMHSQLhssB7mTvIFLpuHwy5u3A6x_wow?e=EhTLFr
Data Description: The collection contains about 17,760 images in 64643 shape (vehicles:8,792 and non-vehicles:8,968), nearly 117MB in size.
Train-test stratified Random Split: 80:20 ratio (14208 training images and 3552 test images)
Models used: Custom MLP ANN, Custom CNN, Pre-trained EfficientNetV2S and Pre-trained MobileNetV3
Best Test set Accuracy: EfficientNetV2S (99.99% accuracy, FP=4, FN=2)
Highest Training time: EfficientNetV2S (18.33 mins)
Lowest Training time: Custom MLP ANN (1.28 mins)