Checklist ONIA 2026 Team Selection Camp (Extended Squad) – Test 1 · Task 3
Video Shazam
Match each 10-second audio embedding to the corresponding video embedding among 20 candidates.
The task
Random 10-second clips were cut from videos; the video tracks were encoded with V-JEPA 2 and the audio tracks with WavLM. About 30,000 aligned audio–video embedding pairs are given for training.
Test data is grouped into samples of 20 audio and 20 video embeddings; for each audio the contestant predicts the index of the matching video. Validation pairs come from the training videos, while the private test also contains clips from unseen videos.
Abridged and translated by SOTA from the official Romanian materials. The official statement has the exact rules, and it wins wherever this summary differs.
At a glance
- You get
main.pyloadingaudio_embeddings(N × 768) andvideo_embeddings(N × 1024) for training, andtest_data: a list of samples withsample_id, 20 audio (768-d) and 20 video (1024-d) embeddings. Archive of 289,639,171 bytes.- You submit
- CSV with header subtaskID,
sample_id, 0 … 19; one row per sample giving the predicted video index for each of the 20 audio clips. - Scoring
- Accuracy; accuracies 0–70 map uniformly to 0–20 points and 70–98 to 20–100 points.
- Rules
- On-site, Timișoara (West University of Timișoara and Politehnica University of Timișoara)
- No internet access; only preinstalled offline resources (NLTK, spaCy
en_core_web_sm, scikit-learn, Gensim glove-wiki-gigaword-50, fastText without pretrained models, and the Hugging Face models distilbert-base-uncased, MoritzLaurer DeBERTa-v3-large-mnli-fever-anli-ling-wanli, roberta-large-mnli, all-MiniLM-L6-v2 and whisper-small) - A GPU Jupyter environment hosted by UVT was provided during the tests
- Problem statements unlocked with a password handed out in the room
- Format
- ONIA 2026 Lot Lărgit, test 1, 23 May 2026.