A tensorflow_2.0 implementation of SSD (Single Shot MultiBox Detector) .
- Python >= 3.6
- TensorFlow == 2.1.0
- numpy == 1.17.0
- opencv-python == 4.1.0.25
- Download the PASCAL VOC 2012 dataset.
- Unzip the file and place it in the 'dataset' folder, make sure the directory is like this :
|——dataset
|——VOCdevkit
|——VOC2012
|——Annotations
|——ImageSets
|——JPEGImages
|——SegmentationClass
|——SegmentationObject
- Run write_voc_to_txt.py to generate voc.txt.
- Run train.py to start training, before that, you can change the value of the parameters in configuration.py.
- Change the test_picture_dir in configuration.py.
- Run test.py to test on single picture.