Exploring object-centric and scene-centric CNN features and their complementarity for human rights violations recognition in images.
Grigorios KalliatakisShoaib EhsanAles LeonardisKlaus D. McDonald-MaierPublished in: CoRR (2018)
Keyphrases
- object models
- image features
- static images
- target object
- visual appearance
- objects in cluttered scenes
- image regions
- images depicting
- occluded objects
- multiple objects
- recognizing objects
- multiple images
- individual objects
- object recognition
- real world scenes
- complex scenes
- geometric information
- image set
- man made objects
- image segments
- object representation
- wire frame
- object model
- three dimensional
- varying illumination conditions
- object features
- background clutter
- input image
- visual scene
- partial occlusion
- test images
- point features
- acquired images
- feature correspondences
- three dimensional objects
- spatial relationships
- textural information
- d objects
- viewing angle
- intensity images
- real world objects
- scene images
- convolutional neural network
- object parts
- geometric constraints
- feature extraction
- salient objects
- real objects
- bounding box
- action recognition
- spatial configuration
- object classes
- image classification
- keypoints
- d scene
- photometric properties
- text localization and recognition
- spatial location
- spatial information
- moving objects
- viewing direction
- object class
- cad model
- object segmentation
- lighting conditions
- image matching
- range finder
- object detection
- image retrieval
- image sequences