Speridlabs releases ENEAS, a text-promptable tracking method claiming gains over SAM3
joecole · x · 2026-09-08
Speridlabs published ENEAS (Embedding-guided Neural Ensemble for Adaptive Segmentation), a unified text-promptable method for instance tracking and semantic discovery.
- Inputs/outputs: a natural-language prompt plus ordered video frames or unordered image collections; returns segmentation masks.
- Instance tracking: name one instance and ENEAS follows it through occlusion, extreme scale changes, and frame exits/re-entries without drifting to look-alikes (statues, paintings, reflections).
- Semantic discovery: name a category and it finds every instance, excluding doppelgangers.
- vs SAM3: the authors claim it outperforms SAM3 on temporal hallucinations, occlusions, spatial fragmentation, and semantic misclassification.
Paper and code are available on arXiv and GitHub.
More from Multimodal
- Running MiniMax H3 video gen on 4GB VRAM: blurry hands at 512p or 6+ hours per clip at native res — unbenannt1 · 2026-09-08
- Creator ships weekly AI series with InVideo agents, keeping creative control for free — LudovicCreator · 2026-09-08
- Grok-Generated Fake 2000s Seoul DV Home Video Goes Viral for Being Undetectable — SimplyAnnisa · 2026-09-08
- Sketch first, let ChatGPT paint over: a workflow where its image model shines — flowersslop · 2026-09-08
- Open-source GPT-6 Astra + Seedance 2.5 pipeline swaps characters in any video — matchaman11 · 2026-09-08
- MIT Multimodal AI Course 'How to AI (Almost) Anything' Videos Released Free — pliang279 · 2026-09-08