By Cognivio Team
This repository contains our submission for the Data Mining Competition held during Hology 8 Universitas Brawijaya (2025) in the qualifying round.
We experimented with several approaches using different model architectures and backbone combinations to improve performance. The scenarios we explored include:
- Training a model from scratch
- Using CSR-Net as the backbone
- Using VGG as the backbone
- Combining SFCN + VGG (backbone)
Among these, the SFCN + VGG configuration achieved the best results.
You can find the corresponding notebook here:
notebooks/pretrained-model/vgg/sfcn-160-epochs.ipynb
In the qualifying round, our team ranked 28th out of 50 participants.
You can view the complete scoreboard in this Google Spreadsheet:
Competition Scoreboard