scripts

Various script to support Traffix development procceses.

Machine Learning

Reproducibility

1. Download COCO dataset
2. preproceses using the respective script
3. Model ready to be trained

prep_dataset_1

This script download COCO train and val data, and filter COCO classes dataset.

filter.py

This script inteded to extract/filter any selected COCO classes.

prep_dataset_2_&_train

This script adjust the COCO train and val filtered annotations dataset to adjust the default classes number to the numbered of filtered classes.

Result

Final Dataset: https://drive.google.com/file/d/1OW9g67a5n__nufpveFrv_5Yck1HOLy3I/view?usp=sharing

{
classes:
    bicycle: 0
    car: 1
    motorcycle: 2
    bus: 3
    truck: 4
}

Trained Model 116 epoch: https://github.com/Traffix-C23-PC636/scripts/releases/download/v0.1.0-alpha/best.pt

epoch	train/box_loss	train/cls_loss	train/dfl_loss	metrics/precision(B)	metrics/recall(B)	metrics/mAP50(B)	metrics/mAP50-95(B)	val/box_loss	val/cls_loss	val/dfl_loss	lr/pg0	lr/pg1	lr/pg2
116(latest trained epoch)	1.1649	1.1808	1.3005	0.63305	0.50011	0.54295	0.34452	1.3287	1.2983	1.3942	0.002575	0.002575	0.002575