Relationship between occlusion and num_points_in_box

seungjun_Lee · February 1, 2020, 5:42am

Hi,

I wonder what is the specific criterion which divides dataset by the level of Occlusion.

For example, what is the specific number of threshold for each occlusion rate?
if num_lidar_points(annotation_box, pc) > threshold :
annotation_box[‘occlusion’] = ‘some what level of occlusion’

Thanks,

Lee.

holger-motional · February 1, 2020, 6:05am

Hi,

we don’t have a field called occlusion in nuScenes. There is only one in KITTI and we just said a dummy value when exporting our data in KITTI format. What we do have is visibility, which is annotated by humans: https://github.com/nutonomy/nuscenes-devkit/blob/master/schema.md#visibility

seungjun_Lee · February 1, 2020, 9:09am

Thank you for the quick reply!

Sorry that I mistakenly said visibility to occlusion.

So does " What we do have is visibility, which is annotated by humans " mean that there is no really a specific criterion on visibility level?

so just annotate the visibility level of 0-40%, 40-60%, 60-80% and 80-100% by some kind of human intuition?

holger-motional · February 2, 2020, 2:37am

“The visibility of an instance is the fraction of annotation visible in all 6 images. Binned into 4 bins 0-40%, 40-60%, 60-80% and 80-100%.”

github.com

nutonomy/nuscenes-devkit/blob/master/schema.md#visibility

Database schema
==========
This document describes the database schema used in nuScenes.
All annotations and meta data (including calibration, maps, vehicle coordinates etc.) are covered in a relational database.
The database tables are listed below.
Every row can be identified by its unique primary key `token`.
Foreign keys such as `sample_token` may be used to link to the `token` of the table `sample`.
Please refer to the [tutorial](https://www.nuscenes.org/tutorial) for an introduction to the most important database tables.


category
---------

Taxonomy of object categories (e.g. vehicle, human). 
Subcategories are delineated by a period (e.g. human.pedestrian.adult).
```
category {
   "token":                   <str> -- Unique record identifier.
   "name":                    <str> -- Category name. Subcategories indicated by period.
   "description":             <str> -- Category description.

This file has been truncated. show original

“The visibility attribute specifies the percentage of object pixels visible in the panoramic view of all cameras.”

github.com

nutonomy/nuscenes-devkit/blob/master/instructions.md

# nuScenes Annotator Instructions 

# Overview
- [Instructions](#instructions)
- [Special Rules](#special-rules)
- [Labels](#labels)
- [Attributes](#attributes)
- [Detailed Instructions and Examples](#detailed-instructions-and-examples) 

# Instructions
+ Draw 3D bounding boxes around all objects from the [labels](#labels) list, and label them according to the instructions below. 
+ **Do not** apply more than one box to a single object.
+ Check every cuboid in every frame, to make sure all points are inside the cuboid and **look reasonable in the image view**.
+ For nighttime or rainy scenes, annotate objects as if these are daytime or normal weather scenes.

# Special Rules 
+ **Minimum number of points** : 
    + Label any target object containing **at least 1 LIDAR or RADAR point**, as long as you can be reasonably sure you know the location and shape of the object. Use your best judgment on correct cuboid position, sizing, and heading. 
+ **Cuboid Sizing** : 
    + **Cuboids must be very tight.** Draw the cuboid as close as possible to the edge of the object without excluding any LIDAR points. There should be almost no visible space between the cuboid border and the closest point on the object.

This file has been truncated. show original