gramuah/pose-estimation-study

The challenge of simultaneous object detection and pose estimation: a comparative study

The challenge of simultaneous object detection and pose estimation: a comparative study

By Daniel Oñoro-Rubio, Roberto J. López-Sastre, Carolina Redondo-Cabrera and Pedro Gil-Jiménez.

This is a repository with the original implementation of the object detection and pose estimation solutions described in our IMAVIS journal paper.

License

This repository is released under the MIT License (refer to the LICENSE file for details).

Citing

If you make use of this data and software, please cite the following reference in any publications:

@Article{Onoro-Rubio2018,
author     = {O\~noro-Rubio, D. and L\'opez-Sastre, R.~J. and Redondo-Cabrera, C. and Gil-Jim\'enez, P.},
title   = {The challenge of simultaneous object detection and pose estimation: a comparative study},
journal = {IMAVIS},
year    = {2018},
volume = {79},
pages = {109-122},
issn = {0262-8856},
doi = {https://doi.org/10.1016/j.imavis.2018.09.013},
}

Contents

1. Requirements: software
2. Requirements: hardware
3. Installation
4. Pre-trained models
5. Usage
6. Demo: test

Requirements: software

1. Requirements for Caffe and pycaffe (see: Caffe installation instructions)

Note: Caffe must be built with support for Python layers!

# In your Makefile.config, make sure to have this line uncommented
WITH_PYTHON_LAYER := 1
# It's also recommended that you use CUDNN
USE_CUDNN := 1

You can download our Makefile.config for reference.

2. Python packages you also need: cython, python-opencv, easydict

Requirements: hardware

1. For training, the models are required a GPU with at least 6 GB of memory and CUDA support.

Installation

Step 1: Clone this repository

git clone https://github.com/gramuah/pose-estimation-study.git

Step 2: We'll call the directory that you cloned PROJECT_ROOT

Step 3: Build the Cython modules

```Shell
cd $PROJECT_ROOT/lib
make
```

Step 4: Build Caffe and pycaffe

```Shell
cd $PROJECT_ROOT/caffe-fast-rcnn
# Now follow the Caffe installation instructions here:
#   http://caffe.berkeleyvision.org/installation.html

# If you're experienced with Caffe and have all of the requirements installed
# and your Makefile.config in place, then simply do:
make -j8 && make pycaffe
```

Step 5: Downloading datasets

In order to test or to train any of the models included in this repository, it is needed one of the following datasets:

• PASCAL3D+: http://cvgl.stanford.edu/projects/pascal3d.html
• ObjectNet3D: http://cvgl.stanford.edu/projects/objectnet3d

The datasets must be manually downloaded from the author's platform and placed in the data directory.

Download pre-trained models

Direct links to download the pre-trained models can be accessed in the file data/scripts/README.md.

Usage

This repository includes the code needed to perform a complete training and testing of any of the proposed models (i.e.: Single-path, Specific-path, and Specific-network) by using the PASCAL3D+ or the ObjectNet3D. All the running scripts are located in: experiments/scripts/<experiment>.sh.

As an example, to train and test our Single-path on the ObjectNet3D, just execute:

```Shell
cd $PROJECT_ROOT
./experiments/scripts/objectnet_single-path.sh [GPU_ID]
# GPU_ID is the GPU you want to train on
```

The output is written underneath $PROJECT_ROOT/output.

Trained models are saved under:

```
output/<experiment directory>/<dataset name>/
```

Test outputs are saved under:

```
output/<experiment directory>/<dataset name>/<network snapshot name>/
```

Demo: test

We provide the pre-trained models of our SPECIFIC-NETWORK model, hence it is possible to run the test.

To run the test on the Pascal3DPlus just execute:

```Shell
cd $PROJECT_ROOT
./experiments/scripts/demo_pascal_3D_network-specific [GPU_ID]
# GPU_ID is the GPU you want to train on
```

Or to run it for the ObjectNet3D, execute:

```Shell
cd $PROJECT_ROOT
./experiments/scripts/demo_objectnet_network-specific.sh [GPU_ID]
# GPU_ID is the GPU you want to train on
```