SSD-gluon

• gluon implementation of SSD in the paper "SSD: Single Shot MultiBox Detector"[Paper].
• This code borrowed from https://zh.gluon.ai/chapter_computer-vision/ssd.html.

Install Required Packages

First ensure that you have installed the following required packages:

• gluon. gluon is the interface of mxnet. The version of mxnet is mxnet0.12.0, maybe other version is ok.
• Opencv (instructions). Here is opencv-2.4.9.

Datasets

• In the implementation of the SSD, we will use PASCAL VOC2007 and VOC2012 datasets. Run scripts/make_DetIter/prepare_pascal.sh to build Record data for detection. You may change im2rec.py a little, i.e. input str, then transform to int. We only modify resize, quality, num-thread, for example:

if type(args.resize) == str:
        args.resize = int(args.resize)

Training and Validating a Model

• Run the following script to train and validate the model.

python trainval_ssd.py

You could change some arguments in the trainval_ssd.py, like num_epochs, gpus.

Testing a Model

• Run the following script to test the model.

python predict.py