HM3D-Sem: Meta AI's Dataset Of Semantically-Annotated 3D Indoor Spaces
Meta AI has revealed what seems to be an updated version of the Habitat Matterport 3D Semantics Dataset, nearly doubling the number of scenes from the original.
Created on October 12|Last edited on October 12
Comment
The Habitat Matterport Dataset is a dataset of 1000 high-resolution 3D scans of indoor environments, such as houses or offices. With habitat datasets like it, AI researchers can develop robots for use in real-world environments in the safety and efficiency of simulated spaces.
An offshoot dataset, called the Habitat-Matterport 3D Semantics Dataset, takes a handful of the main dataset's scenes and annotates them with object labels and identifying color textures. In v0.1 of HM3D-Sem, announced last March, 120 scenes were annotated to an average of around 650 objects per scene thanks to 12,000+ hours of human effort.
Today, Meta AI has released what seems to be an updated version of HM3D-Sem with even more annotated scenes and higher object counts. Though, the webpage linked from the tweet at the time of writing still holds information on the v0.1 dataset, a reply tweet details that this updated dataset now includes 216 scenes, close to doubling the original count.
The linked webpage is still displaying v0.1 information and the GitHub link on the page still points to here, which seems to contain all regular HM3D dataset downloads. The Habitat Matterport datasets must be downloaded from the Matterport website, requiring an account and a request to access the datasets.
Find out more
Add a comment
Tags: ML News
Iterate on AI agents and models faster. Try Weights & Biases today.