3D point-cloud labeling
Techniques for 3D point cloud labeling are described. An example of labeling includes initializing a point-cloud environment according to a configuration; loading of point-cloud data into a memory buffer of a device, wherein the point-cloud data is compatible with a virtual reality (VR) environment and a non-VR environment; drawing of at least the loaded point-cloud data into a point-cloud environment of the device; receiving user input in a task user interface of the device; and in response to the user input, performing an operation to one or more of a bounding box and quality of a label.
1. A computer-implemented method comprising:
receiving a configuration for a point-cloud frame labeling service;
initializing a point-cloud environment in a virtual reality (VR) device according to the configuration;
loading point-cloud data and 3D bounding boxes into a memory buffer of the VR device, wherein the point-cloud data is compatible with a VR environment and a non-VR environment;
drawing the loaded point-cloud data and 3D cuboid annotations by the VR device;
receiving user input in a task user interface of the VR device; and
in response to the input, performing an operation to a bounding box.
2. The computer-implemented method of claim 1 , further comprising converting point-cloud data from a first format into a second format that is compatible with virtual reality-based browsers and computer-based browsers.
3. The computer-implemented method of claim 2 , wherein the second format comprises:
coordinate system information;
virtual reality transform and origin information; and
a point-cloud vertex list.
4. A computer-implemented method comprising:
initializing a point-cloud environment according to a configuration;
loading point-cloud data into a memory buffer of a device, wherein the point-cloud data is compatible with a virtual reality (VR) environment and a non-VR environment;
drawing of at least the loaded point-cloud data into a point-cloud environment of the device;
receiving user input in a task user interface of the device; and
in response to the user input, performing an operation to one or more of a bounding box and quality of a label.
5. The computer-implemented method of claim 4 , further comprising:
converting point-cloud data from a first format into a second format that is compatible with virtual reality-based browsers and computer-based browsers.
6. The computer-implemented method of claim 5 , wherein the second format comprises:
coordinate system information;
virtual reality transform and origin information; and
a point-cloud vertex list.
7. The computer-implemented method of claim 4 , wherein the initializing a point-cloud environment comprises:
launching one or more jobs in a machine learning model training service to provide an interface between point-cloud data stored by a provider network and the device.
8. The computer-implemented method of claim 4 , further comprising:
loading 3D bounding boxes into a memory buffer; and
drawing 3D cuboid annotations for the loaded 3D-bounding boxes.
9. The computer-implemented method of claim 4 , wherein the device is a virtual reality headset.
10. The computer-implemented method of claim 9 , wherein the input is a captured hand gesture to create a bounding box and the task user interface is to render the bounding box including changing a color for point-cloud data to reflect the bounding box.
11. The computer-implemented method of claim 9 , wherein the input is a captured hand gesture to update a bounding box and the task user interface is to render the updated bounding box including changing a color for point-cloud data to reflect the updated bounding box.
12. The computer-implemented method of claim 9 , wherein the input is a captured hand gesture to update a point-cloud data point and the task user interface is to change a color for point-cloud data to reflect the updated point-cloud data point.
13. The computer-implemented method of claim 9 , wherein the input is a captured hand gesture to receive feedback and the task user interface is to determine a quality for a bounding box associated with the captured hand gesture, wherein the feedback is generated based on labeled data or from a machine learning model.
14. The computer-implemented method of claim 4 , wherein the input is a label quality indication.
15. The computer-implemented method of claim 4 , wherein edges of rendered bounding boxes have a values for red, green, blue, and transparency.
16. A system comprising:
a first one or more computing devices to implement a storage service in a multi-tenant provider network; and
a second one or more computing devices to implement a point-cloud data labeling service in the multi-tenant provider network, the point-cloud data labeling service including instructions that upon execution cause the point-cloud data labeling service to:
initialize a point-cloud environment according to a configuration;
load point-cloud data from the storage service into a memory buffer of a device, wherein the point-cloud data is compatible with a virtual reality (VR) environment and a non-VR environment;
draw of at least the loaded point-cloud data into a point-cloud environment of the device;
receive user input in a task user interface of the device; and
in response to the user input, perform an operation to one or more of a bounding box and quality of a label.
17. The system of claim 16 , wherein the point-cloud data is in a format that comprises:
coordinate system information;
virtual reality transform and origin information; and
a point-cloud vertex list.
18. The system of claim 16 , wherein edges of rendered bounding boxes have a values for red, green, blue, and transparency.
19. The system of claim 16 , further comprising:
a virtual reality device, wherein the device is the virtual reality device.
20. The system of claim 16 , wherein the device is a desktop computer.