lloydjie1/EfficientDet-tf

EfficientDet Implementation using TensorFlow 2

Python

EfficientDet

EfficientDet implementation for Object Detection using Tensorflow2

Generate Anchor (Optional)

Generate optimal anchor for your dataset using Anchor Optimizer
Change the values of ratios and scales attribute of AnchorParameters class in utils/util.py

Train

cd project directory
Run python setup.py build_ext --inplace
Change classes variable in utils\config.py based on your dataset
Run python train.py for training using imagenet pretrained weights
After training is finished, change weight_path variable value to trained weight path, example weight_path=weights/D4/model1_0.1.h5
Run python train.py for final training

Test

Run python test.py

Dataset structure

├── Dataset folder 
    ├── IMAGES
        ├── 1111.jpg
        ├── 2222.jpg
    ├── LABELS
        ├── 1111.xml
        ├── 2222.xml
    ├── train.txt
    ├── val.txt

Note

xml file should be in PascalVOC format
for making train.txt and val.txt, see VOC2012/ImageSets/Main/train.txt

Reference