Code repository for visualizing and manipulating HandNet
Branch: master
Clone or download
Fetching latest commit…
Cannot retrieve the latest commit at this time.
Permalink
Type Name Latest commit message Commit time
Failed to load latest commit information.
CreateH5Data.m
DisplaySession.m
Indices.mat
Parameters.mat
README.md
ShowResults.m
compute_maP.m
derotate.m
generateList.m
rotateAround.m
rotateData.m

README.md

HandNet

Code repository for visualizing and manipulating HandNet.

Current project page is here. Data needs to be downloaded from here (into respective folders):

The DerotGT versions were generated by applying the derotate function on the ground truth orientation of each example and then using rotateData. This was then applied as an oracle to create the "ground truth derotated" training and validation data. The Perturbed versions were generated by randomly rotating (uniformly between [-90,90]) the original data around the camera focal point in both 2D (in-plane) and 3D (around the z-axis) also using the rotateData function. The reason for this was to ensure that the validation and test data were not similar to any of the training data. Perturbing the data in this way provides a very strong test of generalization of any learning method on the original data. The perturbed versions of our results are those used in in our BMVC'15 paper. They can be downloaded here:

We dont recommend using the the non-perturbed version of the results but for the sake of completeness we provide our results here:

Explore the results

Run the script ShowResults.m after downloading the results data. This script can also easily be used to benchmark your own results.

Visualize HandNet data

For Matlab or Octave users just run DisplaySession.m to explore the data and change the desired visualization accordingly.

Current examples dont include Python. However the data is in MAT format and reading it with Python is straightforward:

For example (assuming you downloaded the validation data to Validation folder):

import scipy.io
data = scipy.io.loadmat('Validation/Data_0000000.mat')

Citation

Please cite this paper ([LINK] (http://arxiv.org/pdf/1507.05726.pdf)) if you use HandNet in your work.

@article{wetzlerBMVC15,
  Author = {Wetzler, Aaron and Slossberg, Ron and Kimmel, Ron},
  Journal = {British Machine Vision Conference},
  Title = {Rule of Thumb: Deep Derotation for Improved Fingertip Detection,},
  Year = {2015}
}