Market1501 Person Re-Identification Task Using Relational Networks
./src folder contains a couple of source files for the implementation.
./src/model/ReidRN.py contains a relation network which calculates a similarity of two image/feature pair input.
./src/model/ReidRNSelf.py contains a relation network which calculate a feature for a single image/feature input.
./src/data/dataset_loader.py contains loader classes for images or pre-extracted features.
./src/utility/eval_metrics.py contains evaluation function for Market1501.
./src/trainer/extract_features.py can be used to extract features from images using Resnet.
./src/data/market1501_rn.py contains the dataset class.
./src/data/samplers.py contains batch samplers. Pairwise sampling is used to learn the relations between different identities.
./src/trainer/train_rn.py is the script for training and testing, using image pairs to extract relations.
./src/trainer/train_rn_self.py is the script for training and testing, using a single image to extract the relations.
./src/utility/utils.py contains some utility functions.
./src/utility/losses.py contains different types of loss function classes.
This work is inspired by the work of Santoro et al.