RepNet-Vehicle-ReID

Vehicle re-identification implementing RepNet

Vehicle ReID task:

Basic principle for vehicle ReID task:

Using a two-branch deep convolutional network to project raw vehicle images into an Euclidean space where distance can be directly used to measure the similarity of arbitrary two vehicles. For simplicity, triplet loss or coupled cluster loss is replaced here by arc loss which is widely used in face recognition.

Test result

Network structure:

Reference:

Deep Relative Distance Learning: Tell the Difference Between Similar Vehicles
Learning a repression network for precise vehicle search

Dataset:

VehicleID dataset

Pre-trained model

model
extract code: 62wn