For detailed description how to train your own detection model using a custom dataset and evaluate it read the Medium story:
This project is using Conda for project environment management.
Setup the project environment:
$ conda env create -f environment.yml
$ conda activate detectron2-licenseplates
or update the environment if you git pull the repo previously:
$ conda env update -f environment.yml
To launch end-to-end license plates detection training with Faster R-CNN ResNet-50 backbone on 2 GPUs, one should execute:
$ python train.py --config-file configs/lp_faster_rcnn_R_50_FPN_3x.yaml --num-gpus 2
To train the model with RetinaNet ResNet-50 backbone run:
$ python train.py --config-file configs/lp_retinanet_R_50_FPN_3x.yaml --num-gpus 2
Model evaluation is done at the and of the training but you can run it alone:
$ python train.py --config-file configs/lp_faster_rcnn_R_50_FPN_3x.yaml --eval-only MODEL.WEIGHTS output/model_final.pth
or
$ python train.py --config-file configs/lp_retinanet_R_50_FPN_3x.yaml --eval-only MODEL.WEIGHTS output/model_final.pth
To execute prediction on some sample data from test dataset with Faster R-CNN ResNet-50 backbone (which is default), run:
$ python predict.py --config-file configs/lp_faster_rcnn_R_50_FPN_3x.yaml MODEL.WEIGHTS output/model_final.pth
or
$ python predict.py --config-file configs/lp_retinanet_R_50_FPN_3x.yaml MODEL.WEIGHTS output/model_final.pth
to run prediction on RetinaNet
React
Typescript
Django
Laravel
Facebook
Microsoft
Google
Alibaba
D3
Tencent