SLIDE 1
Rotational Rectification Network (R2N): Enabling Pedestrian Detection for Mobile Vision
Xinshuo Weng1, Shangxuan Wu1, Fares Beainy2, Kris M. Kitani1
1Carnegie Mellon University, 2Volvo Construction Equipment
Rotational Rectification Network (R2N): Enabling Pedestrian - - PowerPoint PPT Presentation
Rotational Rectification Network (R2N): Enabling Pedestrian - - PowerPoint PPT Presentation
Rotational Rectification Network (R2N): Enabling Pedestrian Detection for Mobile Vision Xinshuo Weng 1 , Shangxuan Wu 1 , Fares Beainy 2 , Kris M. Kitani 1 1 Carnegie Mellon University, 2 Volvo Construction Equipment WACV 2018, Lake Tahoe
SLIDE 2
SLIDE 3
Pedestrian Detection
- Results on Caltech dataset
Zhang et al. Is Faster R-CNN Doing Well for Pedestrian Detection? ECCV, 2016.
SLIDE 4
Arbitrary-Oriented Pedestrian Detection
SLIDE 5
Arbitrary-Oriented Pedestrian Detection
SLIDE 6
Arbitrary-Oriented Pedestrian Detection
- Random failure cases on Caltech dataset.
SLIDE 7
Why is it interesting?
Imagine the cases:
- Mobile phones
SLIDE 8
Why is it interesting?
Imagine the cases:
- Mobile phones
- UAVs/drones
SLIDE 9
Why is it interesting?
Imagine the cases:
- Mobile phones
- UAVs/drones
- Construction vehicles on a
rugged terrain
SLIDE 10
Why is it interesting?
Imagine the cases:
- Mobile phones
- UAVs/drones
- Construction vehicles on a
rugged terrain
- Wearable cameras
- ….
SLIDE 11
Why is it interesting?
Imagine the cases:
- Mobile phones
- UAVs/drones
- Construction vehicles on a rugged terrain
- Wearable cameras
- ….
Camera orientation can be very flexible with respect to the ground in the real world.
SLIDE 12
Modeling Rotation Invariance or Equivariance
SLIDE 13
Modelling Rotation Invariance/Equivariance
Rotating the inputs
- Data augmentation
- TI-Pooling [Laptev et al
CVPR’ 16]
- ….
- Cons:
○ Low efficiency ○ More parameters
Rotating the filters Changing sampling grids
SLIDE 14
Modelling Rotation Invariance/Equivariance
Rotating the inputs
- Data augmentation
- TI-Pooling [Laptev et al,
CVPR’ 16]
- ….
- Cons:
○ Low efficiency ○ More parameters
Rotating the filters
- RotEqNet [Marcos et al,
ICCV’ 17]
- ORNs [Zhou et al, CVPR’
17]
- ….
- Cons:
○ Approximated rotations ○ Memory issues
Changing sampling grids
SLIDE 15
Modelling Rotation Invariance/Equivariance
Rotating the inputs
- Data augmentation
- TI-Pooling [Laptev et al,
CVPR’ 16]
- ….
- Cons:
○ Low efficiency ○ More parameters
Rotating the filters
- RotEqNet [Marcos et al,
ICCV’ 17]
- ORNs [Zhou et al, CVPR’
17]
- ….
- Cons:
○ Approximated rotations ○ Memory issues
Changing sampling grids
- Spatial Transformer
[Jaderberg et al, NIPS’ 15]
- Deformable ConvNets [Dai
et al, ICCV’ 17]
- GPPooling (Ours)
- ….
SLIDE 16
Global Polar Pooling (GPPooling)
Inputs Activations
SLIDE 17
GPPooling vs Pooling
GPPooling Pooling
Noh et al. Learning Deconvolution Network for Semantic Segmentation? ICCV, 2015.
SLIDE 18
What is Rotational Rectification Network (R2N)?
R2N = Rotation Estimation Module (including GPPooling) + Spatial Transformer
SLIDE 19
Results
SLIDE 20
Take Home Messages
- GPPooling can be used to model global rotation equivariance/invariance in
general CNNs.
- R2N is easy to plug in and improves the performance on oriented detection