A Tensorflow implementation of SSD from the 2016 paper by Wei Liu. As a classical network framework of one-stage detectors, SSD are widely used. Our code is based on balancap/SSD-Tensorflow. The official and original Caffe code can be found in Caffe.
You can edit the data and path information yourself in the tf_convert_data.py file, then run python tf_convert_data.py
Note the previous command generated a collection of TF-Records instead of a single file in order to ease shuffling during training.
SSD300 trained on VOC0712balancap/SSD-Tensorflow
python train.py You can track your training on the tensorboard real time
In the CITY data set, single-class car have reached the 84% mAP
We implemented Mobilenet2-SSD, you can change framework in nets/ssd_300_mobilenet2.py Mobilenet-v2 is an improved version of Mobilenet, but we found that it's not a big improvement for detection.
There are two improved network structures for SSD, CEBNet ICME2019, and FFBNet ICIP2019.