Home

Awesome

Single View Metrology in the Wild

Code and data for Single View Metrology in the Wild, Zhu et al, ECCV 2020

To be released. Stay tuned by watching (subscribing to) the repo from the button on the upper right corner.

Installation

conda create -y -n scalenet python=3.6
conda activate scalenet
pip install -r requirements.txt
conda install nb_conda

cd maskrcnn-benchmark
conda install cudatoolkit
python setup.py build develop
cd ..

git clone https://github.com/NVIDIA/apex.git
cd apex
python setup.py install --cuda_ext --cpp_ext # if you see an error about commenting out an IF setence, do it
cd ..

python setup_maskrcnn_rui.py build develop

with Jupyter notebook

Lanuch jupyter notebook. Kernel -> Change Kernel [to scalenet]

Todolist

Notes

Camera Calibration Network

Location

./RELEASE_ScaleNet_minimal

Demo

./RELEASE_ScaleNet_minimal/demo-evalCameraCalib-SUN360-RELEASE.ipynb

Description

This network is trained on SUN360 dataset with supervision of some camera parameters (e.g. roll, pitch, field of view (or equivalently focal length), which can be converted to horizon as well). The release model takes in a random image, and estimates:

Note that geometric relationships exist between those items. Specifically:

Scale Estimation Inference on COCOScale

Preparation

Location

./RELEASE_ScaleNet_minimal

Demo

./RELEASE_ScaleNet_minimal/demo-evalScaleNet-COCOScale-RELEASE.ipynb

Description

The demo loaded images from the COCOScla dataset and runs inference and visualization of the scale estimation task.