SteerX: Creating Any Camera-Free 3D and 4D Scenes with Geometric Steering
Fuente:
arXiv
Saved in:
| Main Authors: | , , , , , |
|---|---|
| Format: | Preprint |
| Published: |
2025
|
| Subjects: | |
| Online Access: | |
| Tags: |
Add Tag
No Tags, Be the first to tag this record!
|
| _version_ | 1866909711940452352 |
|---|---|
| author | Park, Byeongjun Go, Hyojun Nam, Hyelin Kim, Byung-Hoon Chung, Hyungjin Kim, Changick |
| author_facet | Park, Byeongjun Go, Hyojun Nam, Hyelin Kim, Byung-Hoon Chung, Hyungjin Kim, Changick |
| contents | Recent progress in 3D/4D scene generation emphasizes the importance of physical alignment throughout video generation and scene reconstruction. However, existing methods improve the alignment separately at each stage, making it difficult to manage subtle misalignments arising from another stage. Here, we present SteerX, a zero-shot inference-time steering method that unifies scene reconstruction into the generation process, tilting data distributions toward better geometric alignment. To this end, we introduce two geometric reward functions for 3D/4D scene generation by using pose-free feed-forward scene reconstruction models. Through extensive experiments, we demonstrate the effectiveness of SteerX in improving 3D/4D scene generation. |
| format | Preprint |
| id |
arxiv_https___arxiv_org_abs_2503_12024 |
| institution | arXiv |
| publishDate | 2025 |
| record_format | arxiv |
| spellingShingle | SteerX: Creating Any Camera-Free 3D and 4D Scenes with Geometric Steering Park, Byeongjun Go, Hyojun Nam, Hyelin Kim, Byung-Hoon Chung, Hyungjin Kim, Changick Computer Vision and Pattern Recognition Recent progress in 3D/4D scene generation emphasizes the importance of physical alignment throughout video generation and scene reconstruction. However, existing methods improve the alignment separately at each stage, making it difficult to manage subtle misalignments arising from another stage. Here, we present SteerX, a zero-shot inference-time steering method that unifies scene reconstruction into the generation process, tilting data distributions toward better geometric alignment. To this end, we introduce two geometric reward functions for 3D/4D scene generation by using pose-free feed-forward scene reconstruction models. Through extensive experiments, we demonstrate the effectiveness of SteerX in improving 3D/4D scene generation. |
| title | SteerX: Creating Any Camera-Free 3D and 4D Scenes with Geometric Steering |
| topic | Computer Vision and Pattern Recognition |
| url | https://arxiv.org/abs/2503.12024 |