Export

While Tables are designed to be used directly in model training, one might want to export the modified data into a common format such as CSV or COCO, or to a custom format. This can be achieved by using the built-in exporters in the CLI and Python API, the 3lc-tools package or custom iteration.

Built-in exporters

Built-in exporters can be used through the Python API or CLI, and currently CSV, COCO and JSON are supported output formats.

Using Table.export()

To export a 3LC Table, you can use the table.export() method from the tlc package to export.

import tlc

table = tlc.Table.from_url("path/to/3LC/Table")
table.export("path/to/output/folder/my_dataset.csv")

The example above will export a 3LC Table to a CSV file named my_dataset.csv in the output location you specified.

If you want to export a 3LC Table containing bounding boxes to the COCO format, you just need to use “json” as the file’s extension such as my_dataset.json.

Using the 3LC CLI

In a terminal (with the tlc Python Package installed), the command line tool can be invoked as follows:

3lc export path/to/table.json <output-path>

The output format will be deduced from the extension of <output-path> and the contents of the table, but can also be explicitly specified using the --format option. For more details, use

3lc export --help

In-memory export

To export the data in a tlc.Table to a Pandas DataFrame, use

df_from_table = table.to_pandas()

This will return the rows of the Table in a Pandas dataframe.

Custom format

To export to a custom format, iterate over the tlc.Table directly and write the data to your desired format. Here we assume you have a function my_export_fn that performs the necessary steps to serialize and write the data.

table = Table.from_url(input_url)

my_table_dict = {"col1": [], "col2": []}
for sample in table:
    my_table_dict["col1"] = sample[0]
    my_table_dict["col2"] = sample[1]

my_export_fn(my_table_dict)

Exporting a Table to YOLO format using the tlc_tools package

To export a 3LC Table or a set of Tables (such as train and val Tables) containing bounding boxes to the YOLO format, you will need another designated method in the tlc_tools package, because it will write out a number of files for images, labels, and the YAML file instead of a single file. The tlc_tools Python package provides a set of ready-to-use methods and functionalities at your disposal. You can follow the instructions in our public GitHub repository to install it.

import tlc
from tlc_tools.yolo_export import export_to_yolo

train_table = tlc.Table.from_url("path/to/train/Table")
val_table = tlc.Table.from_url("path/to/val/Table")

export_to_yolo(
    tables={
        "train": train_table,
        "val": val_table,
    }, 
    output_url="path/to/output/folder", 
    dataset_name="my_dataset", 
    image_strategy="copy",
)

In the export_to_yolo method, you can specify both the train and val Tables at once or just one single Table. The value of the dataset_name will be the YAML file’s name. There are several options for the argument image_strategy. The default is copy, which will copy all the images into the output folder. If you don’t want to copy the images, you can specify it as ignore.