Research
Amap releases ABot-Recon for long-video 3D reconstruction
Alibaba's mapping business has released a reconstruction model and implementation intended for extended video sequences.
Weights, licenses, evaluations and what it really takes to deploy a model on your own terms.
Official resource ↗Alibaba's mapping business has released a reconstruction model and implementation intended for extended video sequences.
The model combines text and visual inputs, with weights released through Hugging Face according to TechNode.
Public weights and an architecture study detail Qwen's sparse attention, gated residual design and separate n-gram embedding capacity.
SenseTime's August U1.5 Lite preview combines visual understanding, 4K generation and localized editing, with both downloadable weights and a hosted image API.