Skip to content

An object detection system that detects the category and location of an object that is being pointed at by the user's hand, based on the attention map calculated from a image-classification neural network..

Notifications You must be signed in to change notification settings

daohanlu/Attention-Map-Point-to-tell

Repository files navigation

Weakly Supervised Point-to-tell

An object detection system that detects the category and location of an object that is being pointed at by a (blind) user's hand, based on the attention map calculated from a image-classification neural network (hence weak supervision, since training images only need a categorical label).

Summary Paper is in summary paper.pdf
This work is built upon Point-to-Tell-and-Touch

Main programs are under cnn/Guided-Attention-Inference-Network. crop_hands.py crops hands from a video by color masking and saves them to a folder as pngs with metadata such as the location of the fingertip. gt_transfer.py reads the VOC SBD dataset and overlays the png images of hands on top of objects of a class of interest.

Machine Learning code based on code by github user @alokwhitewolf and paper Tell Me Where to Look: Guided Attention Inference Network

About

An object detection system that detects the category and location of an object that is being pointed at by the user's hand, based on the attention map calculated from a image-classification neural network..

Resources

Stars

Watchers

Forks

Releases

No releases published

Packages

No packages published