6M-param ZipDepth model matches 350M-param depth model at 10-15x speed, mobile-ready demo
Bedrovelsen · x · 2026-09-11
Developer pablovelagomez1 released a Hugging Face Space (rerun + Gradio integration) for ZipDepth-PromptDA: a 6M-parameter depth model distilled from the 350M-parameter Prompt Depth Anything, running 10-15x faster and feasible on mobile. Workflow: scan a room with Polycam, then upscale depth with promptda / zip-promptda.
More from Research
- VikingRAG matches top RAG accuracy using just 5-33% of tokens via directory-aware retrieval — _reachsumit · 2026-09-11
- GLIE regenerates full page embeddings from 4 vectors, slashing visual document retrieval storage — _reachsumit · 2026-09-11
- MotionSplicer at ECCV 2026: Remaking 4D Video Motion Part by Part — drsrinathsridha · 2026-09-11
- NCP-ArchPreview report: latent-space LMs via next concept prediction — NCP Team · 2026-09-11
- Recursive Code World Models builds complex 3D worlds from a single image as executable code — Zhiqi Li · 2026-09-11
- CARDEA: chain-of-box reasoning makes coronary angiography AI diagnoses auditable — Jia-Jen Lee · 2026-09-11