Combine knowledge of many vision transformers into a smaller multi-talented model

Improve segmentation in bin picking by using part awareness

Get 3D layout of room from a panoramic image with LGT-Net

Convert a blurry image to a sharp video using a continuous intensity function with E-CIR

Survey of continuous human action recognition

Survey of re-identifying people seen by multiple cameras for various problem types

Real-time object segmentation in video with faster training using EfficientVIS

Better recognition of human actions using two-stage detection transformers

Improve scene segmentation with smaller models by distilling knowledge with SKR+PEA

Estimate the pixel map homograhy using image features and pixels

Segment objects of any type in image using model trained without manual annotations with FreeSOLO

Better object detection when domain change with balanced domains using BD-DIR

Improve object discovery with self-supervised transformers using TokenCut

Better image stitching by using depth information

Improve any image feature matcher on large appearance changes using OETR

Human pose estimation with 80% smaller model and 68% less CPU using STNet

Improve vision models by pretrain on uncurated images without supervision with SEER

Better 3D surface reconstruction from point clouds by using sensor's viewpoint

Remove haze in a single image using estimated transmission map with EDN-GTM

Improved object detection when test and train domains differ with MS-DAYOLO