Can you make Machine Learning models to Retrieve Landmarks?

Google is organizing the 4th iteration of its Landmark Retrieval Challenge on Kaggle- The Google Landmark Retrieval 2021 Challenge. With the help of this completion, Google aims to leverage the Machine Learning practitioners community on Kaggle to help retrieve images that have the same landmark as that in the queried image. Retrieving images that are similar to a given image is one of the key problems in Computer Vision, and has uses in searching based on images.

The Google Landmark Retrieval 2021 Challenge Demonstration Image.
Image by Varun Kulkarni from Pixabay. Source

The Competition

In this competition, you will be provided with a training dataset that will consist of images and the landmarks that are present in the images. You will have to train a model that will retrieve images that contain the same landmark as the query image. Along with the training dataset, the query images will be provided in the test folder and the images on which retrieving will be performed will be provided in the index folder.

The solutions will be evaluated by the Mean Average Precision @ 100, also known as the mAP@100 metric. The run-time for the entire notebook, irrespective of whether a CPU is used or a GPU, should not exceed 12 hours. While the use of TPU will not be allowed during the submissions, they can still be used for training the models.

This competition is being held in conjunction with Landmark Recognition Challenge.


The competition began on 10th August 2021. And while the final submission deadline is on 1st October 2021, the deadline for entry into the competition and team merger is 24th September 2021. So far, there are already 80 competitors from 59 teams who have made 387 entries.


Since this is a competition, there are obviously going to be prizes. Top 3 position holders will be a Kaggle merchandise of their choice. Although it may not sound like much as compared to other competitions on Kaggle, there is also another component to the prize.

The competition is a part of the Instance Level Recognition Workshop at the International Conference on Computer Vision, ICCV 2021. Competitors at the top of the leaderboard will be invited to give a talk at the workshop. Consequently, the winners of the completion will also have to submit their papers explaining their methods by 8th October 2021. The paper should at least be 2 pages long in the double-column format.

To be a part of the Google Landmark Retrieval 2021 challenge, you just need to have a Kaggle Account. To know more about this challenge and participate in it, click here.

Good Luck!

Other Challenges

Kaggle, a subsidiary of Google is by far the biggest community of Data Science and Machine Learning practitioners. Kaggle regularly organizes competitions in various domains and on behalf of various organizations.

You can check out the blog on the Kaggle CHAII Research Challenge by Google, click here.

You can check out the blog on the Kaggle Gravitational Wave Detection challenge by G2Net, click here.

You can check out the blog on the Kaggle Volatility Forecasting challenge by Optiver, click here.

Read more here: Source link