This repo contains some prototype tools for video annotations. These single-page HTML tools are designed to be used for Amazon Mechanical Turks crowd-sourcing annotation.
Some demo frames/GIFs are dumped from Charades dataset.
Contributers: @ranjaykrishna and me
Provided with a reference video clip and the interested human action, the task is to draw a bounding box around an object involved in the action.
See bbox.html.
Provided with a reference video clip and the interested human action, the task is to label all relationships between the actor and the object.
Find all occurring actions in each short video clip.
See multiclass.html