Video reconstruction using vp detection results

Used sources

• (cvpr'22)VanishingPoint_HoughTransform_GaussianSphere
• (neurVPS)NeurVPS

How to use

• set enviroment

cd cvpr22
conda create -y -n vpd
conda activate vpd
conda env update --file config/environment.yml

• download models data
  download

• make dirs

  • cvpr22
    • saved_results
      • nyu
      • scannet
      • su3
  • neurVPS
    • saved_results
      • scannet
      • su3
      • tmm17

• modify cvpr code in video_predict.py and predict.py

  • img_size: video_predict.py 35 line
  • weight_name: predict.py 123 line

• run python code

cd cvpr22 # or cd neurVPS
python video_predict.py nyu #scannet su3

Specification

name	info
os	ubuntu18.04
cpu	Intel(R) Core(TM) i7-6700K CPU @ 4.00GHz
gpu	GeForce RTX 2080 Ti Rev. A
ram	16G

Libraries

name	version
python	3.7.0
pytorch	1.12.0-cuda11.3-cudnn8
numpy	1.21.2
matplotlib	3.2.2
opencv-python	4.8.0.74
scipy	1.7.3
scikit-image	0.18.3

Metrics

• Performance

  name	sec per frame(avg)
  NYU(cvpr)	0.20 sec
  ScanNet(cvpr)	0.17 sec
  SU3(cvpr)	0.18 sec
  SU3(neur)	1.09 sec
  ScanNet(neur)	1.20 sec
  TMM17(neur)	0.85 sec

• AA(Angular Accuracy) Graph
  • cvpr
    
  • neur
    
  • all
    

SnapShot

• cvpr NYU Dataset Result
  

Reference paper

(cvpr) https://arxiv.org/abs/2203.08586
(neurVPS) https://arxiv.org/abs/1910.06316