Add support for the Eiffel Tower underwater dataset

[clementinboittiaux]

Hi @skydes,

We recently published a dataset for long-term visual localization in underwater scenarios called Eiffel Tower. The dataset embeds four different visits of the same hydrothermal vent between 2015 and 2020. It is available at https://www.seanoe.org/data/00810/92226/.

The associated paper describing the dataset will be published shortly. To put in simply, the dataset was created using the following pipeline:

• Models are first built independently for each year with the help of the vehicle's navigation data.
• They are then registered in a common reference frame using TEASER++ (https://github.com/MIT-SPARK/TEASER-plusplus) and ICP.
• A global model embedding images of all visits is then built with the help of individual models that are now in the same reference frame.

The whole code was tested on Ubuntu 20.04.

Feel free to ask if you have any questions.

[clementinboittiaux]

Hi @skydes,
The data paper is now available on arxiv: https://arxiv.org/abs/2305.05301
Don't hesitate if you have any questions!

[clementinboittiaux]

Hi @skydes,

The data paper has been published on IJRR website: https://doi.org/10.1177/02783649231177322
Do you think it would be ok to merge?

Best,
Clémentin