The challenge is closed.
In this competition, you are challenged to build a deep learning model that identifies different types of natural scenes in a dataset of images, where the images have been noisy (between 15-25% of the image was randomly covered by a black rectangle).
The dataset counts with 6 natural scene categories: buildings, forest, glacier, mountain , sea , street
This is an individual competition, only for SC 2021 participants.
You will work with the dataset available at: dataset_sc_2021.zip
Your task is to propose a model to obtain best classification results using the provided test subset (a part of the data_scg_2021.zip).
The participants of this challenge will be required to send the code of the created solution (in a ZIP file) for further check.
The results will be evaluated by the organizers and three best entries will be awarded.
The winners of the competition will be asked to send a short video (1-5 minutes) or PDF slides with a presentation of their solution. The film or slides will be placed on the SCG 2021 website, therefore matters relating to the privacy of submitters must be properly considered when preparing the video).
Submissions will be evaluated on a proposal presentation and a weighted F1 score using a scikit-learn function.
other parameters - default.
The proposal presentation will hold 3-5min and is about your progress, proposal, or just ideas about dealing with the problem.
Also, we will require you to submit two files with your results. The submission website will be announced on Slack.
We require a .csv file where for each id in the test set, you must predict a type of natural scene (or label). The file should contain a header and have the following format:
B indicates buildings
F indicates forest
G indicates glacier
M indicates mountain
S indicates sea
ST indicate street
The submission file should be named:
2. We require you to send the code of the created solution (in a ZIP file) for further check. Use the following format for the name
Sources and Materials
We provided two templates (in tensorflow and pytorch) for a quick start.
Information about data set
The dataset used in this challenge is the "Intel Image Classification" dataset provided by Puneet Bansal.
Source: Intel Image Classification
Licence: CC BY-SA 4.0
The dataset has been processed for the needs of this challenge