ITLPCampus Dataset Reader

A module that implements a PyTorch dataset reader for the ITLPCampus dataset.

More details about the dataset can be found on the ITLP-Campus Dataset page.

Below is a usage example of the dataset reader that is implemented as part of the OpenPlaceRecognition library.

Usage example

You should start with initializing the dataset reader by creating an instance of the opr.datasets.itlp.ITLPCampus class with appropriate parameters.

from opr.datasets.itlp import ITLPCampus
from pathlib import Path
from torch.utils.data import DataLoader

# Path to your ITLP dataset directory
track_dir = Path("/path/to/ITLP_Campus_outdoor/00_2023-02-21")  # outdoor track example

# Initialize the dataset reader
dataset = ITLPCampus(
    dataset_root=track_dir,                      # track directory
    sensors=["front_cam", "back_cam", "lidar"],  # sensors to load
    load_semantics=True,                         # load semantic masks
    mink_quantization_size=0.5,                  # point cloud voxelization size
    max_point_distance=None,                     # max distance to keep points (None for no limit)
    indoor=False,                                # specify if it's an indoor track
)

# Access a single sample from the dataset
sample = dataset[0]  # returns a dictionary with data from the first frame

The sample will contain a dictionary with the following keys:

  • "idx": index of the sample in the dataset

  • "pose": 7-element pose vector [tx, ty, tz, qx, qy, qz, qw]

  • "image_front_cam": tensor of front camera image (if “front_cam” in sensors)

  • "image_back_cam": tensor of back camera image (if “back_cam” in sensors)

  • "mask_front_cam": tensor of front camera semantic mask (if load_semantics=True)

  • "mask_back_cam": tensor of back camera semantic mask (if load_semantics=True)

  • "pointcloud_lidar_coords": tensor of LiDAR point coordinates (if “lidar” in sensors)

  • "pointcloud_lidar_feats": tensor of LiDAR point features (if “lidar” in sensors)

To create a DataLoader for batch training or inference:

# Create a DataLoader using the dataset's collate_fn
batch_size = 4
dataloader = DataLoader(
    dataset,
    batch_size=batch_size,
    shuffle=True,
    collate_fn=dataset.collate_fn,
    num_workers=4
)

# Iterate through the data batches
for batch in dataloader:
    # Process batch data
    # batch["images_front_cam"] # [batch_size, 3, H, W]
    # batch["images_back_cam"]  # [batch_size, 3, H, W]
    # batch["masks_front_cam"]  # [batch_size, 1, H, W]
    # batch["poses"]            # [batch_size, 7]
    # batch["pointclouds_lidar_coords"] # ME batched coordinates format
    # batch["pointclouds_lidar_feats"]  # ME batched features format
    pass

For indoor datasets, you can specify train/test splits:

# Split floors for indoor dataset
train_split = ["1", "2", "3"]  # floor numbers for training
test_split = ["4", "5"]        # floor numbers for testing

# Create training dataset
train_dataset = ITLPCampus(
    dataset_root=track_dir,
    subset="train",              # "train", "val", or "test"
    indoor=True,
    train_split=train_split,
    test_split=test_split
)

# Create test dataset
test_dataset = ITLPCampus(
    dataset_root=track_dir,
    subset="test",
    indoor=True,
    train_split=train_split,
    test_split=test_split
)

More examples of using the ITLPCampus dataset reader can be found in the notebooks: