search

OpenRobotLab / EmbodiedScan

[CVPR 2024] EmbodiedScan: A Holistic Multi-Modal 3D Perception Suite Towards Embodied AI
https://tai-wang.github.io/embodiedscan/
Apache License 2.0
394 stars 26 forks source link

[Docs] What type of data in Matterport3D do we need to download? #12

Closed RongkunYang closed 4 months ago

RongkunYang commented 4 months ago

Branch

main branch https://mmdetection3d.readthedocs.io/en/latest/

📚 The doc issue

The full matterport3D dataset is very large, may I ask what type of data do we need to download?

Suggest a potential alternative/fix

No response

mxh1999 commented 4 months ago

For running, you will need the matterport_color_images and matterport_depth_images. These are necessary inputs for the process.

For visualization, you can use the matterport_mesh, house_segmentations, and region_segmentations. These provide visual representations.

For debugging, you can download the cameras, matterport_camera_poses, and matterport_camera_poses. Note that the annotation files already include extrinsic and intrinsic information, but it can be convenient to download them separately for debugging purposes.

If you have limited space, the model can still work properly with just the matterport_color_images and matterport_depth_images.

RongkunYang commented 4 months ago

OK, thank you for your fast response, I'll try it.

jjlinghu commented 4 months ago

Hi! If I just want to debug, do I also have to download these three datasets and convert them in addition to EmbodiedScan.

mxh1999 commented 4 months ago

Hi! If I just want to debug, do I also have to download these three datasets and convert them in addition to EmbodiedScan.

Yes.

Specifically, if you want to debug a scene, you can download only the data related to this scene.