Nodar GroundTruth (NGT) gives ground-truth-level accuracy based on time-synchronized images you provide. In this document, we outline the expected input to NGT, the configuration required in your AWS ecosystem, and the output you can expect from us.
We expect the images you take to be time-synchronized. Image Synchronization Requirements shows the degree of synchronization required for different scenarios. We can generally tolerate synchronization <50μs, for highly dynamic environments, we may require <10μs.
In your S3 bucket, you will need to organize each dataset according to the Nodar format. The format is shown in Input Format page along with sample intrinsics and extrinsics files. Optionally, if you have a special rectified view that you want to assume full control, you can supply a reprojection_config.ini file.
The file intrinsics.ini will contain the intrinsic parameters of each camera. Similarly, the file extrinsics.ini will store the relative pose between the cameras described using Euler angles and translation vector.
You will need to grant our regional principal role access to your S3 bucket. You can find instructions to add a bucket policy here: https://docs.aws.amazon.com/AmazonS3/latest/userguide/add-bucket-policy.html
Here is an example bucket policy. You can copy it to your json bucket policy and add your bucket name and principal.
{
"Version": "2012-10-17",
"Statement": [
{
"Sid": "granting-access-to-the-prod-account",
"Effect": "Allow",
"Principal": {
"AWS": "arn:aws:iam::812508052848:role/NodarSaasStack-prod-us-ea-NodarPrincipalRole6B88118-1cPZAoMAI8KA"
},
"Action": "s3:*",
"Resource": [
"arn:aws:s3:::your-s3-bucket-name",
"arn:aws:s3:::your-s3-bucket-path/*"
]
}
]
}
The format of the deliverables is in Output Format .
All outputs are saved back in the directory provided in the S3 path and prefixed with gt-.
The point cloud soup is composed of disparity images, the rectified left image, and metadata stored in gt-details folder. These are sufficient to recreate the point cloud in post processing and may save you on data storage.
Disparity images are saved as a 32bit, 4 decimal fixed-point in tiff format.