Zero human labels: auto-generated soccer tracking dataset hits 62.5 HOTA, beating fine-tuned FairMOT
RexDouglass · x · 2026-10-09
Developer Alex Bodner shares progress on an auto-labeled player-tracking dataset: given three words, a model generates the full labeled tracking dataset with no human annotation anywhere. Using Astra + SAM3 to generate fragmentary tracks and distilling them into cheaper detection/tracking models, the pipeline scores 62.5 HOTA on SoccerNet's 46-clip unseen test split — well above label-trained baselines like fine-tuned FairMOT (57.9), ByteTrack (47.2), and DeepSORT (36.7). They switched from expensive Astra to Sol (GPT-6.1 Sol), which performs about the same at roughly a quarter of the cost. Still behind the 2023 challenge winners, but self-improvement is working.
More from Research
- Ex-self-driving ML engineer writes long-form on the practice of semi-supervision — Visual_Ability · 2026-10-09
- RLVR misses 'all minimal correct answers' problems; new credit assignment doubles finds — thoma_gu · 2026-10-09
- Debunked: AI did not solve the Millennium Prize Navier-Stokes problem — gerardsans · 2026-10-09
- New open-source 3JSBench evaluates LLMs on generating coherent Three.js 3D assets — ycombinator · 2026-10-09
- Moonworks' Lunara: Sub-10B Diffusion Mixture Transformer Tops Aesthetic and Human Blind Evaluations — paper-crow · 2026-10-09
- Study: Local domains take 41.4% of AI citations in Brazil, 38.3% in UK across ChatGPT and Gemini — gaganghotra_ · 2026-10-09