PVNet: Pixel-wise Voting Network for 6DoF Pose Estimation - cm107 Fork

Navigation

• clean-pvnet's original README.md

Requirements

• CUDA 10.X
• Python 3.7+
• Pip
• virtualenv
• Recommended OS: Linux (tested)

Installation

Before installation, please install the following apt packages.

sudo apt-get install libglfw3-dev libglfw3

If you have not already done so, activate the virtual environment that you want to install this package to. If you haven't created a virtual environment yet, please do so now.

virtualenv venv
source venv/bin/activate

After your virtual environment is activated, you can install this package.

git clone https://github.com/cm107/clean-pvnet.git
cd clean-pvnet
pip install -e .

Training

Training-related code has not been refactored yet, and thus is highly dependent on relative paths, softlinks, and settings that are hard-coded in the original config. For now, please refer to clean-pvnet's original README.md. This document will be updated once the training portion of this repository is refactored.

Inference

As of right now, the inference code has been completely refactored and re-organized into the clean_pvnet package.

The important inference-related classes and methods are listed below:

• PnpDrawSettings - Drawing settings for PVNet model inference
• PnpPrediction - Contains Pnp prediction data for a single detection (i.e. a single bbox)
• PnpPredictionList - A list of Pnp prediction data
• PVNetFrameResult - A collection of all pnp data and metadata in a single frame
  • draw - Draws pvnet prediction on input image given settings defined in PnpDrawSettings
• PVNetFrameResultList - All pnp predictions and metadata in all frames
  • save_to_path - Saves all prediction data to a dump file
  • load_from_path - Loads all prediction data from a dump file
• PVNetInferer - Worker class that handles all pvnet inference
  • predict - Gets PVNetFrameResult from an input image
  • infer_coco_dataset - Runs inference systematically on a COCO dataset

For details on what parameters can be used in the above classes and methods, refer to clean_pvnet/infer/pvnet_inferer.py. For a simple example of how to use all of these classes together in an inference script, refer to test/example_inference.py