跳到正文
HuggingFace Daily Papers(社區熱門論文)·· 2026-10-15AI 評分34

Gen4P:影片生成器是 4D 感知器

Gen4P: Video Generators are 4D Perceivers

AI 導讀

Gen4P 將影片生成器改作前饋式 4D 感知模型,採用單次編碼、按需解碼以支援空間靈活性,並以連續時間嵌入支援時間靈活性。它以凍結的深度 LoRA 作為可插拔的幾何先驗,並在 flamingo 等具挑戰性的場景中保留結構及追蹤動作。

正文

TL;DR: Gen4P repurposes a video generator for feed-forward 4D perception, with single-pass encoding, on-demand decoding for spatial flexibility, continuous-time embeddings for temporal flexibility, and a frozen depth LoRA as a plug-in geometry prior.

As a result, Gen4P achieves faithful structure preservation and motion tracking on challenging scenes like flamingo.

Please refer to https://gen4p.github.io/ for more results.

來源:HuggingFace Daily Papers(社區熱門論文) · huggingface.co