Single-Shot-Detector

Implementation of Single Shot Detector in tensorflow.

Files

• base_layer.py : Contains the vgg16 architecture as the base layer for feature extraction
• ssd300.py : Contains the layers following the base layer, which output the predictions.
• loss.py : Contains the loss function of the network.
• batch_gen.py : Contain functions for parsing the training files and generating the training set.
• utils.py : Contains encoding and decoding funtions. The encoding function encodes the training label in a format similar to network output. The decoding function turns back the network output in the label format.
• train.py : Performs the training of the model.
• predict.py : Tests the trained model.

Training is done on PASCAL VOC dataset.