Skip to content
Go to file

Deep Reflectance Codes (DRC), A Hashing Approach for Material Retrieval

Created by Hang Zhang, Kristin Dana and Ko Nishino


This repository contains the code for reproducing the results in the paper (ECCV 2016):

     Author = {Hang Zhang and Kristin Dana and Ko Nishino},
     Title = {Friction from Reflectance: Deep Reflectance Codes for Predicting Physical Surface Properties from One-Shot In-Field Reflectance},
     Booktitle = {Proceedings of the European Conference on Computer Vision ({ECCV})},
     Year = {2016}

If you use the code for your research, please cite our paper. Link to the project

Get Started

  • The code relies on vlfeat, and matconvnet, which should be downloaded and built before running the experiments. (Supprting the version matconvnet-1.0-beta18.) You can use the following commend to download them:

      git clone --recurse-submodules
  • Download the model VGG-M to data/models (older models can also be updated using the vl_simplenn_tidy function).

  • Download the following dataset to data/

    • Reflectance Disks (reflectance)
    • Flickr Material Database (fmd)
    • Describable Textures Dataset (dtd)
    • Textures under varying Illumination (kth)
  • run_Experiments.m reproducing general material recogniton results

  • HashForFriction_Demo.m reproducting friction prediction results

General Material Recogniton Results

reflectance 64.5% 51.9% 60.1% 58.8% 59.9% 60.2%
FMD 48.3% 65.0% 59.4% 57.7% 64.8% 65.5%
DTD 43.6% 52.6% 52.3% 53.1% 55.4% 55.8%
KTH 72.0% 73.7% 75.6% 54.4% 76.6% 77.2%


This work was supported by National Science Foundation award IIS-1421134 to KD and KN and Office of Naval Research grant N00014-16-1-2158 (N00014-14-1-0316) to KN.

We thank vlfeat and matconvnet teams for creating and maintaining these excellent packages. We would like to thank Felix Yu for hashing algorithms and Cimpoi for FV-CNN encoders. The copyrights of original code reserve.


Deep Reflectance Codes (DRC): A Hashing Approach for Material Retrieval



You can’t perform that action at this time.