Introduction

This is the code our ICPR2018 paper "DeepFirearm: Learning Discriminative Feature Representation for Fine-grained Firearm Retrieval".

How to run the code

Version info

The code is written using the PyTorch version 0.3.0. So In order to run this code, you may install version 0.3.0 of PyTorch or adapt it to the newer version of PyTorch.

Instructions for downloading dataset

You can download the dataset from here. Two separate data are used for the experiment. One is for classification training, and the other is for the retrieval training. After downloading this dataset, extract it under the folder data using the following commands:

tar -zxvf firearm-train-val.tar.gz -C data/ # for the classification data

and

tar -zxvf firearm-dataset.tar.gz -C data/ # for the retrieval data

Train the classification model

In order to train the classification model, run the following command:

python train_cls.py

Train the retrieval model after classification

To get better retrieval performance, we further fine-tune the model using retrieval task based on the classification model. To train the model, use the following command:

python train_retr_from_cls.py

Benchmark on test set

To check the model's performance on test set, run the following command:

python benchmark_on_test.py

It will show both the mAP and rank-k accuracy for different feature dimensions.

Citation information

If you use this dataset or use our code, please cite the following work:

@INPROCEEDINGS{HJD2018DFLD,
author={J. Hao and J. Dong and W. Wang and T. Tan},
booktitle={2018 24th International Conference on Pattern Recognition (ICPR)},
title={DeepFirearm: Learning Discriminative Feature Representation for Fine-grained Firearm Retrieval},
year={2018},
volume={},
number={},
pages={3335-3340},
keywords={feature extraction;feedforward neural nets;image classification;image representation;image retrieval;learning (artificial intelligence);convolutional neural networks;single margin contrastive loss;firearm images;double margin contrastive loss;negative image pairs;positive image pairs;fine-grained recognition;Firearm 14k;image retrieval techniques;social media;fine-grained Firearm retrieval;discriminative feature representation;Training;Task analysis;Labeling;Correlation;Image retrieval;Forensics;Convolutional neural networks},
doi={10.1109/ICPR.2018.8545529},
ISSN={1051-4651},
month={Aug},}