adataset

Convert dataset

adataset is used to convert datasets (nuScenes, KITTI, ApolloScape) to Apollo record file. This way we can guarantee the consistency of training data and test data, including sensor intrinsics and extrinsics parameter files, thus speeding up model validation.

Install

pip3 install adataset

Usage

We first introduce the use of the command, and then introduce how to use the dataset with adataset.

Command options

The options for adataset command are as follows:

• --dataset(-d) Choose the dataset, support list n, k, a, w, means "n:nuScenes, k:KITTI, a:ApolloScape, w:Waymo"
• --input(-i) Set the dataset input directory.
• --output(-o) Set the output directory, default is the current directory.
• --type(-t) Choose conversion type, support list rcd, cal, pcd, means "rcd:record, cal:calibration, pcd:pointcloud", default is rcd.

Convert record files

You can use below command to convert dataset to Apollo record file. For example convert nuScenes dataset in dataset_path to Apollo record. The output default is the current directory, and the type default is rcd.

adataset -d=n -i=dataset_path

The name of the nuScenes record file is scene_token.record, and KITTI is result.record, and ApolloScape is frame_id.record

Convert calibration files

You can use below command to convert dataset to apollo calibration files. There maybe multi sense in one dataset, and we create calibration files for each scene.

adataset -d=n -i=dataset_path -t=cal

Camera intrinsics

Camera intrinsics matrix. ref link

• D: The distortion parameters, size depending on the distortion model. For "plumb_bob", the 5 parameters are: (k1, k2, t1, t2, k3).
• K: Intrinsic camera matrix for the raw (distorted) images.
• R: Rectification matrix (stereo cameras only)
• P: Projection/camera matrix

Convert PCD file

You can use below command to convert dataset lidar pcd to normal pcl file, which can display in visualization tools such as pcl_viewer.

adataset -d=n -i=dataset_lidar_pcd_file -t=pcd

If you do not specify a name, the default name of the pcd file is result.pcd, saved in the current directory.

Dataset introduction

There are differences between the data sets, so introduce them separately.

nuScenes

nuScenes Mini compared with the full amount of data, the Mini data set is relatively small. The nuScenes Mini data set is as follows.

nuScenes-Mini
 -maps
 -samples
 -sweeps
 -v1.0-mini

Then we can use the following command to generate the "record/calibration/pcd" file.

// record
adataset -d=n -i=path/to/nuScenes-Mini
// calibration
adataset -d=n -i=path/to/nuScenes-Mini -t=cal
// pcd
adataset -d=n -i=path/to/nuScenes-Mini/samples/LIDAR_TOP/n015-2018-11-21-19-38-26+0800__LIDAR_TOP__1542801007446751.pcd.bin -t=pcd

KITTI

We use KITTI raw data to generate Apollo record file. Be sure to download [synced+rectified data] but not [unsynced+unrectified data]. Note that the calibration data are in [calibration].

dataset

The KITTI raw data is as follows.

2011_09_26_drive_0015_sync
 -image_00
 -image_01
 -image_02
 -image_03
 -oxts
 -velodyne_points

Then we can use the following command to generate the "record/pcd" file.

// record
adataset -d=k -i=path/to/2011_09_26_drive_0015_sync
// pcd
adataset -d=k -i=path/to/2011_09_26/2011_09_26_drive_0015_sync/velodyne_points/data/0000000113.bin -t=pcd

calibration

The KITTI calibration data is as follows:

2011_09_26
 -calib_cam_to_cam.txt
 -calib_imu_to_velo.txt
 -calib_velo_to_cam.txt

Then we can use the following command to generate the Apollo "calibration" files.

adataset -d=k -i=path/to/2011_09_26 -t=cal

ApolloScape

We use ApolloScape Detection/Tracking dataset to generate record file.The ApolloScape Detection/Tracking data is as follow.

Training data
 -detection_train_pcd_1.zip
 -detection_train_bin_1.zip
 -detection_train_label.zip
 -...
 -tracking_train_pose.zip

Testing data
 -detection_test_pcd_1.zip
 -detection_test_bin_1.zip
 -...
 -tracking_test_pose.zip

Before generating record file, we should organize data folders as follow(use test data for example).

├── tracking_test
    ├── pcd
        ├── result_9048_2_frame
            ├── 2.bin
            ├── 7.bin
            └── ...
        └── ...
    ├── pose
        ├── result_9048_2_frame
            ├── 2_pose.txt
            ├── 7_pose.txt
            └── ...
        └── ...

Then we can use the following command to generate the "record/calibration/pcd" file.

// record
python main.py -d=a -i=tracking_test/ -o=records/ -t=rcd
// calibration
python main.py -d=a -i=tracking_test/ -o=records/ -t=cal
// pcd
python main.py -d=a -i=data_path/data.bin -o=data_path/result.pcd -t=pcd