NSTM targets real-time dynamic NVS with minute-long memory
Jimantha · x · 2026-07-23
BabackE teases NSTM, a method for dynamic novel-view synthesis that aims to let you step into a live 3D broadcast or video call and look around freely.
- The post argues current multi-view camera setups leave hidden regions that must be reconstructed from memory.
- Existing stateful and stateless models supposedly cannot do this in real time over long horizons.
- NSTM claims real-time inference plus minute-long memory, targeting dynamic NVS and live 3D viewing.
More from Research
- Paper argues algorithmic neutrality is impossible because relevance itself is value-laden — CurieuxExplorer · 2026-07-23
- Anchor-Align lifts xArm7 real-robot success from 28% to 54% — Dwip Dalal · 2026-07-23
- SetwiseEvalKit scores document sets, not just single search results — Kailin Jiang · 2026-07-23
- DeepSeek-V4 post-training on Ascend SuperPOD reaches 34.22% MFU — Dongfang Li · 2026-07-23
- Essay says “algorithmic neutrality” is a myth and warns philosophy hires can become corporate ethics theater — CurieuxExplorer · 2026-07-23
- iFlytek-backed robot startup bets on world models, body awareness and less real-robot data — 量子位 · 2026-07-23