/EfficientDet

EfficientDet (Scalable and Efficient Object Detection) implementation in Keras and Tensorflow

Primary LanguagePythonApache License 2.0Apache-2.0

EfficientDet

This is an implementation of EfficientDet for object detection on Keras and Tensorflow. The project is based on fizyr/keras-retinanet and the qubvel/efficientnet. The pretrained EfficientNet weights files are downloaded from Callidior/keras-applications/releases

Thanks for their hard work. This project is released under the Apache License. Please take their licenses into consideration too when use this project.

Train

build dataset (Pascal VOC, other types please refer to fizyr/keras-retinanet)

  • Download VOC2007 and VOC2012, copy all image files from VOC2007 to VOC2012.
  • Append VOC2007 train.txt to VOC2012 trainval.txt.
  • Overwrite VOC2012 val.txt by VOC2007 val.txt.

train

  • STEP1: python3 train.py --snapshot imagenet --phi {0, 1, 2, 3, 4, 5, 6} --freeze-backbone --gpu 0 --random-transform --compute-val-loss --batch-size 32 --steps 1000 pascal datasets/VOC2012 to start training. The init lr is 1e-4.
  • STEP2: python3 train.py --snapshot xxx.h5 --phi {0, 1, 2, 3, 4, 5, 6} --gpu 0 --random-transform --compute-val-loss --batch-size 32 --steps 1000 pascal datasets/VOC2012 to start training when val mAP can not increase during STEP1. The init lr is 1e-5 and decays to 1e-6 when loss stops dropping down.

Evaluate

  • python3 eval/common.py to evaluate by specifying model path there.