{
  "fetched_at": "2026-10-18T13:53:47Z",
  "source_url": "https://www.kaggle.com/competitions/ai-spacetech-hackathon/data",
  "pages": [
    {
      "name": "data-description",
      "content": "This dataset encompasses 10917 SAR image patches across Rotterdam in the train set, offered in single-band georeferenced GRD format that has a VV polarisation band present in GeoTIFF format.\n\nAlongside the SAR data, the dataset includes co-registered collection of Maxar WorldView 2 optical images at a 0.5m resolution, for this of image translation competition. \n\nThere are 122 SAR image patches in the test set.\n\n```\nAOI_11_Rotterdam/\n├── train/\n│       ├── sar\n│   ├── [Files: Georeferenced SAR magnitude]\n│       ├── eo\n│   ├── [Files: Georeferenced RGBNIR]\n├── test/\n│       ├── sar\n│   ├── [Files: Georeferenced SAR magnitude]\n\n```\n",
      "isPublished": false,
      "hasIsPublished": false,
      "postTitle": "",
      "hasPostTitle": false,
      "mimeType": "",
      "hasMimeType": false
    },
    {
      "name": "Description",
      "content": "What are EO and SAR Imagery?\n\nEO (Electro-Optical) images as snapshots taken with a regular camera but from space. These images are straightforward and visually intuitive, showing the Earth in colors and textures we're familiar with. On the other hand, SAR (Synthetic Aperture Radar) images don't rely on daylight or clear skies; instead, they use radar to 'see' through clouds and darkness, providing valuable data no matter the weather or time of day. Utilising the microwave portion of the electromagnetic spectrum allows SAR to penetrate through obstacles like clouds, rain, smoke, and foliage. \n\nHowever, SAR images are not as easy to understand at a glance—they present the world in a way that's unfamiliar and often challenging to interpret. This difference creates a unique opportunity for our hackathon: to bridge the gap between the all-weather, all-time reliability of SAR and the intuitive, easily understandable nature of EO images. Solving this challenge could significantly help in Data Augmentation for training new machine learning models, enhancing our understand of SAR image, and developing new algorithms.\n\n![image](https://github.com/protocolbuffers/protobuf/assets/76432265/39743765-9fcc-4a96-8fc2-8b40fa0e2e3b)\n\nTo solve this problem, this hackathon encourages participants to develop solution to transfer style of SAR into EO Satellite Imagery.\n\n",
      "isPublished": false,
      "hasIsPublished": false,
      "postTitle": "",
      "hasPostTitle": false,
      "mimeType": "",
      "hasMimeType": false
    },
    {
      "name": "Evaluation",
      "content": "\nParticipants can chose their own metric, but they must present a detailed report on why they chose a particular metric in the final presentation.",
      "isPublished": false,
      "hasIsPublished": false,
      "postTitle": "",
      "hasPostTitle": false,
      "mimeType": "",
      "hasMimeType": false
    },
    {
      "name": "rules",
      "content": "1. Registration and Team Formation\nAll participants must register by the specified deadline.\nParticipants can register as individuals or as part of a team.\nTeams must have a minimum of 1 and a maximum of 4 members.\nCross-institutional and cross-disciplinary teams are encouraged.\n\n2. Hackathon Duration\nThe hackathon will start at 10:00 AM on March 16 and end at 10:00 AM on March 17.\n\n3. Project Themes and Requirements\nProjects must align with at least one of the hackathon themes provided.\nUse of open-source libraries and external APIs is allowed, provided they are publicly accessible and free to use.\n\n4. Intellectual Property\nParticipants retain full ownership of their projects and intellectual property.\nBy participating, teams agree to allow the hackathon organizers to use their projects for promotional purposes.\n\n5. Judging Criteria\nProjects will be judged based on completeness, documentation and metrics, and alignment with the hackathon themes.\nThe judges' decision will be final and binding.\n\n6. Code of Conduct\nParticipants are expected to behave professionally and respectfully towards all other participants, mentors, and organizers.\nHarassment of any kind will not be tolerated.\nTeams found violating the code of conduct may be disqualified.\n\n7. Submission Requirements\nAll projects must be submitted through the designated hackathon submission platform (Kaggle).\nSubmissions must include a working prototype, source code, and a detailed documentation.\n\n8. Prizes\nPrizes will be awarded to the top one team.\n\n9. Changes to the Rules\nThe hackathon organizers reserve the right to make changes to the rules and regulations to ensure fairness and the smooth running of the event.\n\n10. Participation Agreement\nBy registering for the hackathon, participants agree to abide by all the rules and decisions of the organizers.",
      "isPublished": false,
      "hasIsPublished": false,
      "postTitle": "",
      "hasPostTitle": false,
      "mimeType": "",
      "hasMimeType": false
    },
    {
      "name": "abstract",
      "content": "\nThe objective of this hackathon is to develop deep learning models capable of Image-to-image translation in SAR2EO (Synthetic Aperture Radar to Electro Optical Imagery) which involves converting Synthetic Aperture Radar images that mimic the appearance and characteristics of Electro-Optical satellite images. \n\nExplore novel approaches to improve human understanding of raw Synthetic Aperture Radar (SAR) images, presenting techniques that transform SAR data into comprehensible Electro-Optical imagery.\n\nTo aid in the creation of these models, we have provided a subset of the Spacenet 6 Dataset containing satellite imagery (pair of SAR and Optical) for the year 2020 over Rotterdam, which is accessible from this link below.\n\nYour task is to build a deep learning model that generates EO from SAR Imagery.\n\n\n*This hackathon is organised by [Entrepreneur First](https://www.joinef.com/) in collaboration with [Speciale Invest](https://www.specialeinvest.com/) and [GalaxEye](https://www.galaxeye.space/).*",
      "isPublished": false,
      "hasIsPublished": false,
      "postTitle": "",
      "hasPostTitle": false,
      "mimeType": "",
      "hasMimeType": false
    },
    {
      "name": "Standard Operating Procedure",
      "content": "- **Hackathon Task**\n  - The core task for the hackathon is to perform image-to-image translation, specifically translating SAR images into Electro Optical (EO) images.\n\n- **Dataset Information**\n  - Participants are provided with a dataset consisting of 10917 image pairs, each comprising Electro Optical (EO) and Synthetic Aperture Radar (SAR) images.\n  - These image pairs are located within the `train` folder, further divided into subfolders named `eo` for Electro Optical images and `sar` for Synthetic Aperture Radar images.\n\n- **Evaluation and Prediction**\n  - Participants are encouraged to select their own metrics for evaluating their models. However, the primary objective is to utilize these models for translating the SAR images in the test dataset to optical images\n  - The test dataset can be found in the `test` folder, which contains a `sar` subfolder. Participants are expected to generate EO images as their output predictions.\n\n- **Presentation and Submission Requirements**\n  - Each team must prepare a presentation detailing the (i) methodology (ii) the performance on the chosen metrics (iii) the graph of the chosen metrics for each epoch of training and the optical images generated from the SAR images in the test dataset\n  - This presentation must be submitted through a specified Google Form whose link is provided below.\n  -  Participants must make their notebook(s) **public** while submission. (For more information on how to make it public, visit this [link](https://www.kaggle.com/discussions/getting-started/58109#1804207))\n  - The name of your notebook **must** be titled with the **team name**.\n\n- **Working Environment**\n  - There are no restrictions on the computational environment; participants may choose to work on Kaggle, Google Colab, local machines, or any other platform of their preference.\n  - Despite the freedom to choose their working environment, all participants must submit their final Jupyter notebooks (.ipynb files) exclusively on the Kaggle platform.\n\n- **Scoring criteria**\n\nThe the solutions will be scored on three primary criteria:\n(i) - The completeness of the solution: The solution should be complete i.e atleast on iteration of the translation should be performed on the test images\n(ii) - Performance measurement: The choice of evaluation metrics, the performance over the metrics for each epoch, a neat depiction of models performance on the validation and test datasets\n(iii) - Presentation/Packaging: The neatness of the presentation as well as the readability of the code\n\nNote: Complexity has the least number of points so it is advised to choose a solutions which can be worked out in the limited amount of time rather than choosing a complex solution which will remain incomplete. However, it is allowed to use existing repositories and models to fine tune them for the given task on the provided dataset \n",
      "isPublished": false,
      "hasIsPublished": false,
      "postTitle": "",
      "hasPostTitle": false,
      "mimeType": "",
      "hasMimeType": false
    }
  ]
}