Survey on improving efficiency of computer vision recogntion using deep learning
Efficiently identify people in image in one stage without separate detection step
Answer question about an image using text in scene to find external knowledge
Camera looking at blank wall can determine number of people and activity
Super-resolution video using non-neighboring frames without frame alignment
More accurate action recognition in videos by using the identities of people
Restore image by removing artifacts superimposed in an unknown manner
Real-time YOLOP detects traffic object, drivable area, and road lane
Find location of image anomalies using neural net trained on simulated anomalies
Better 3D reconstruction without exact model match by using database of patches
Improve thermal IR images using neural net to model physics and influence of scene
Detect anomaly in video by reconstructing motion and frame predicting
Segmentation system marks unknown objects which are then incrementally learned
Segment humans in image after self-supervised training on multiple views
Segmenting objects in images with transformers
Dense depth in crowded dynamic scenes using sparse depth and monocular color images
Get segmented bird's-eye view of a scene from one frontal image
Building detection in high-resolution satellite images
Remove haze in a single image by using three components
Get depth in videos where the objects and the camera move