rohithannapureddy26/gdfl

Graininess-aware Deep Feature Learning for Pedestrian Detection

By Chunze Lin, Jiwen Lu, Gang Wang, Jie Zhou.

Introduction

GDFL is a graininess-aware deep feature learning based detector for pedestrian detection. We exploit fine-grained details into deep convolutional features for robust pedestrian detection. You can use the code to evaluate the model for pedestrian detection task. For more details, please refer to our paper and our poster. The GDFL code is based on the implementation of SSD.

Citing GDFL

Please cite GDFL in your publications if it helps your research:

@inproceedings{lin2018graininess,
  title={Graininess-aware deep feature learning for pedestrian detection},
  author={Lin, Chunze and Lu, Jiwen and Wang, Gang and Zhou, Jie},
  booktitle={Proceedings of the European Conference on Computer Vision (ECCV)},
  pages={732--747},
  year={2018}
}

Installation

Build the code. Please follow Caffe instruction to install all necessary packages and build it.

# Modify Makefile.config according to your Caffe installation.
cp Makefile.config.example Makefile.config
make -j8
# Make sure to include $CAFFE-GDFL_ROOT/python to your PYTHONPATH.
make pycaffe

Data Preparation

Download the Caltech dataset and follow the instructions to extract the annotations and images. Put them in the propor folder.

Evaluation

You can download a pre-trained model on Caltech at here and store it at 'models/Caltech/'. You can evaluate this model on a single image or on the Caltech testing set.

# Make sure to include $CAFFE-GDFL_ROOT/python to your PYTHONPATH.
# if you would like to test the model with a single image, you can do:
python examples/scripts/single_image_detection.py

# if you would like to test with Caltech testing set, you can do:
python examples/scripts/test_caltech.py

The results of the Caltech testing set will be generated in 'output/' and can be evaluated with matlab evaluation code.