Recod.ai/LUC - Scientific Image Forgery Detection

Overview

This competition challenges you to build AI models that can find and mark duplicated sections in scientific images, specifically from biomedical research. The goal is to help protect science by detecting image manipulation like copy-move forgery, which is often used to fake results. Your work could significantly improve how scientific integrity is verified.

Requirements

This is an online competition where submissions must be made through Kaggle Notebooks. During the training phase, your CPU or GPU Notebook must run for 4 hours or less, with internet access disabled. External data and pre-trained models are allowed. For the forecasting phase, the run-time limit increases to 9 hours. Your submission file needs to be named 'submission.csv' or 'submission.parquet' and formatted correctly, predicting 'authentic' for images without forgery or providing run-length encoded masks for forgeries using the provided functions.

Prizes

There's a total prize pool of $55,000. The top five teams will receive cash prizes: $17,500 for 1st place, $12,500 for 2nd place, $11,000 for 3rd place, $8,000 for 4th place, and $6,000 for 5th place. Participants can also earn Kaggle Awards Points and Medals.