Checklist NEOAI 2025 Day 2 · Task 4
Cluster Pictures
Cluster augmented images, each given as a pair of feature matrices, into the 32 original images they came from.
The task
Two sets of arrays, X1 and X2, are given; row i of both arrays describes the same image. All data come from 32 original images, each of which was augmented several times and the augmented versions added to the dataset. The contestant must assign every row to one of the 32 clusters.
If a neural network is used, it must be the supplied EmbNet (a pretrained tiny_vit_5m_224 feature extractor), used without modification and without fine-tuning.
Abridged by SOTA from the official materials. The official statement has the exact rules, and it wins wherever this summary differs.
At a glance
- You get
data_1.npzanddata_2.npz(arrays X1 and X2, one row per image),sample_submission.csv.- You submit
- CSV with columns id (row number) and target (cluster).
- Scoring
- Adjusted Rand index (sklearn
adjusted_rand_score). - Rules
- Only the supplied EmbNet may be used as a neural network; its class must not be changed and it must not be fine-tuned.
- The task must be solved on Kaggle (cloud.ru may not be used); the notebook must be Quick-Saved and its version cited in the submission comment.
- General NEOAI 2025 rules apply: by default no pretrained models or additional data unless the statement allows them; no external human help; on-site participants get one NVIDIA V100, an LLM chat (GPT-4o or similar) and a whitelist of websites.
- Format
- NEOAI 2025 main round, Task 4, Day 2 (8 May 2025, 07:00–13:00 UTC per the schedule). Individual; 20 submissions per day.