FLORIDA dataset

This is the released FLORIDA dataset for our paper An Efficient Semi-Automated Scheme for Infrastructure LiDAR Annotation.

The dataset will be made publicly available after the acceptance of our paper.

Statistics

Dataset Format

• Data format follows SUSTechPOINTS

• To quickly view the dataset, unzip the all data and put them under the data folder of SUSTechPOINTS

SUSTechPOINTS
├── data
│   ├── DATA_20220927_010000
│   │   ├── lidar
│   │   │   ├── XXX.pcd
│   │   ├── label
│   │   │   ├── XXX.json
|   ├── DATA_20220927_011000_2300
│   │   ├── lidar
│   │   │   ├── XXX.pcd
│   │   ├── label
│   │   │   ├── XXX.json
...

• Training sequences

DATA_20220927_010000
DATA_20220927_011000_2300
DATA_20220927_140000_1700
DATA_20220927_012000_1580
DATA_20220927_012500_800
DATA_20220927_161500
DATA_20220927_163000
DATA_20220927_170500_600
DATA_20220927_171000_400
DATA_20220927_164000_100
DATA_20220927_164000

• Testing sequences

DATA_20221122_182000_0
DATA_20221121_230000_900

• To load a label file into the unified 3D box definition: (x, y, z, dx, dy, dz, heading), as defined in OpenPCDet

import json

json_path = '<path to data folder>/<sequence_name>/label/<frame_name>.json'
json_file = open(json_path)
label_json = json.load(json_file)
label_list = []
label_names = []
for label in label_json:
    psr = label['psr']
    label_list.append([psr['position']['x'],
                       psr['position']['y'],
                       psr['position']['z'],
                       psr['scale']['x'],
                       psr['scale']['y'],
                       psr['scale']['z'],
                       psr['rotation']['z']])
    label_names.append(label['obj_type'])

Annotation tool

As an ablation study, we tested our annotation scheme and tool on the publicly available LUMPI dataset. The following screen recordings demonstrate the utilization of SOT and MOT models, as well as batch mode correction and helper functions.

SOT demo

Car

Pedestrian

MOT demo

Car & Pedestrian

Examples of correcting the MOT results

Car

Pedestrian

Operations

Provided by SUSTechPOINTS

• Batch mode adjustment
• Automatic box fitting

Our Contributions

• Annotation propagation up to 100 frames using SOT
• Auto-annotation using MOT
• Pedestrian trajectory smoothing and orentation correction
• Static annotation propagation (for stopped or parked vehicles)
• Vehicle orientation correction
• Merging tracks