/depthpose

Primary LanguagePythonMIT LicenseMIT

Depthpose: Absolute Human Pose Estimation with Depth Prediction Network

This repository contains the trained model for the paper. Parts of the code are based on the MegaDepth repository.

Prerequisites

The code was written in Python 2, using the following packages:

  • torch 0.4.1
  • numpy 1.15.4
  • OpenCV 3

Setup

You will need the MuPoTS-3D dataset downloaded from here. Also, 2D pose detections are needed, which are available from here. The archive must be extracted in to the root MuPoTS-3D folder.

The pretrained model can be downloaded from here. You must extract the contents in the base directory of Dephtpose.

Note, if you want to run the model on your own data, you'll have to install OpenPose. For training the model, commit e59215a219a7328a6b9eb0af14b41c35564d399d was used.

Running the model

After the data and model weights were downloaded, you can run the baseline algorithm like this:

python scripts/baseline.py --model-path models/baseline  --mupots <path to mupots dataset>

You can optionally add the --relative argument to calculate relative errors.

To evaluate the full model:

python scripts/full.py --model-path models/end2end/model_weights.pkl  --mupots <path to mupots dataset>

The --relative switch works here as well.