Welcome to the third Landmark Recognition competition! This year, we have worked to set this up as a code competition and collected a new set of test images.

Have you ever gone through your vacation photos and asked yourself: What was the name of that temple I visited in China? or Who created this monument I saw in France? Landmark recognition can help! This technology can predict landmark labels directly from image pixels, to help people better understand and organize their photo collections. This competition challenges Kagglers to build models that recognize the correct landmark (if any) in a dataset of challenging test images.

Many Kagglers are familiar with image classification challenges like the ImageNet Large Scale Visual Recognition Challenge (ILSVRC), which aims to recognize 1K general object categories. Landmark recognition is a little different from that: it contains a much larger number of classes (there are more than 81K classes in this challenge), and the number of training examples per class may not be very large. Landmark recognition is challenging in its own way.

In the previous editions of this challenge (2018 and 2019), submissions were handled by uploading prediction files to the system. This year's competition is structured in a synchronous rerun format, where participants need to submit their Kaggle notebooks for scoring.

This challenge is organized in conjunction with the Landmark Retrieval Challenge 2020, which was launched June 30, 2020. Both challenges are affiliated with the Instance-Level Recognition workshop in ECCV’20.

This is a Code Competition. Refer to Code Requirements for details.

Notebooks   Host Baseline Example 23 votes · a day ago 日本語 EDA for biginner 1 vote · 2 hours ago Simple reading data and better train.csv 7 votes · a day ago GLRec - Images in the form of long strips 6 votes · 4 hours ago Google Landmarks Feature Extraction 0 votes · 16 hours ago 32 discussion topics   Looking for a teammate 1 reply · 12 hours ago Looking for a Team Megathread 10 replies · 10 hours ago Does query expansion help for recognition-via-retrieval? 0 replies · 7 hours ago Last year's top solutions 13 replies · 7 hours ago Dataset confusion 1 reply · 21 minutes ago     Launch8 days ago  

2 months

Rules Acceptance Deadline

  Close2 months  

85

Teams

90

Competitors

281

Entries

Points

This competition awards standard ranking points

Tiers

This competition counts towards tiers

Tags image data

image datax 1733

data type > image data

computer vision

computer visionx 1053

technique > computer vision

custom metric

custom metric

Automatic Tag

Data Description

In this competition, you are asked to take test images and recognize which landmarks (if any) are depicted in them. The training set is available in the train/ folder, with corresponding landmark labels in train.csv. The test set images are listed in the test/ folder. Each image has a unique id. Since there are a large number of images, each image is placed within three subfolders according to the first three characters of the image id (i.e. image abcdef.jpg is placed in a/b/c/abcdef.jpg).

This is a synchronous rerun code competition. The provided test set is a representative set of files to demonstrate the format of the private test set. When you submit your notebook, Kaggle will rerun your code on the private dataset. Additionally, this competition also has two unique characteristics:

• To facilitate recognition-by-retrieval approaches, the private training set contains only a 100k subset of the total public training set. This 100k subset contains all of the training set images associated with the landmarks in the private test set. You may still attach the full training set as an external data set if you wish.
• Submissions are given 12 hours to run, as compared to the site-wide session limit of 9 hours. While your commit must still finish in the 9 hour limit in order to be eligible to submit, the rerun may take the full 12 hours.

GLDv2

The training data for this competition comes from a cleaned version of the Google Landmarks Dataset v2 (GLDv2), which is available here. Please refer to the paper for more details on the dataset construction and how to use it. See this code example for an example of a pretrained model.

If you make use of this dataset in your research, please consider citing:

"Google Landmarks Dataset v2 - A Large-Scale Benchmark for Instance-Level Recognition and Retrieval", T. Weyand, A. Araujo, B. Cao and J. Sim, Proc. CVPR'20

kaggle competitions download -c landmark-recognition-2020 Use the Kaggle API to download the dataset. https://github.com/Kaggle/kaggle-api Copy API command to clipboardKaggle API installation and documentation

Data Explorer

98.27 GB arrow_right folder

test

  arrow_right folder

train

    calendar_view_week

sample_submission.csv

    calendar_view_week

train.csv

 

Summary

arrow_right folder

1.59m files

arrow_right calendar_view_week

4 columns

get_appDownload All chevron_left  

sample_submission.csv(292.99 KB)

get_app   fullscreen   DetailCompactColumn

2 of 2 columns

keyboard_arrow_down     text_formatidsort text_formatlandmarkssort

10345

unique values

1

unique value 00084cdf8f600d00 137790 0.1 000b15b043eb8cf0 137790 0.1 0011a52f9b948fd2 137790 0.1 00141b8a5a729084 137790 0.1 0018aa4b92532b77 137790 0.1 001baaaab791d8db 137790 0.1 002bc16418269c3d 137790 0.1 0043158ee252596c 137790 0.1 00441c9ba2d68ce4 137790 0.1 0044d82ea7654ece 137790 0.1 004f3afc1377b6fb 137790 0.1 004fba3dc75efae7 137790 0.1 0065fe4125634e9c 137790 0.1 006c3aaa4f75b6c4 137790 0.1 00779bb62b0a3387 137790 0.1 0087f2ac45a2244f 137790 0.1 地址：https://www.kaggle.com/c/landmark-recognition-2020/data