SSD_tensorflow

This is a pure tf 1.4.0 implemention of SSD: Single Shot MultiBox Detector. Modified according to the Keras version https://github.com/rykov8/ssd_keras .

Dependency：

python3.5
tensorflow 1.4
opencv2
h5py (optional if you delete one code block in ssd.py)
pickle
numpy

How to predict:

python3 ssd.py --mode=eval --images_dir=path/to/images --label_file=gt_pascal.pkl --checkpoint=path/to/checkpoint

*Parameters are all optional. The default values are in ssd_utils.py

How to train:

python3 ssd.py --mode=train --images_dir=path/to/images --label_file=gt_pascal.pkl --checkpoint=path/to/checkpoint

*Parameters are all optional. The default values are in ssd_utils.py

How to prepare data:

Two ways: 1.Format your data as the Pascal VOC format and run: python3 tools/get_data_from_XML.py You may need to modify some path in tools/get_data_from_XML.py

2.Run tools/get_data_from_txt.py if your data is txt files. Path modification is also needed.

3.Remember to run tools/check_gt_pascal.py to see whether gt_pascal.pkl is correct.