Research map / Video / Video segmentation and tracking
Seeing without Pixels: Perception from Camera Trajectories
Zihui Xue, Kristen Grauman, Dima Damen, Andrew Zisserman, Tengda Han
Read the paperCheck your idea against 137,217 papers
AI-extracted. The measured effects below were read by AI from the paper's full text (no summary yet). Numbers and claims can be wrong or missing context, so check the paper before you cite or build on it.
Measured effects reported in the paper
| What was changed | Where | Measured | Effect |
|---|---|---|---|
| use relative pose sequences instead of absolute (3D+4D) (ablation) | CamFormer · Ego-Exo4D text retrieval | top-1 accuracy (%): 34.74 -> 43.66 | Improved |
| use midpoint reference frame instead of previous pose (ablation) | CamFormer · Ego-Exo4D text retrieval | top-1 accuracy (%): 43.66 -> 44.02 | Improved |
| use 6D rotation instead of 4D quaternion (absolute) (ablation) | CamFormer · Ego-Exo4D text retrieval | top-1 accuracy (%): 34.74 -> 37.82 | Improved |
| use 6D rotation instead of 4D quaternion (relative midpoint) (ablation) | CamFormer · Ego-Exo4D text retrieval | top-1 accuracy (%): 44.02 -> 44.12 | No clear effect |
| add gravity direction input (ablation) | CamFormer · Ego-Exo4D text retrieval | top-1 accuracy (%): 44.12 -> 44.81 | Improved |
| add rotation to absolute translation (ablation) | CamFormer · Ego-Exo4D text retrieval | top-1 accuracy (%): 32.84 -> 37.82 | Improved |
| use Aria poses instead of video-estimated pi3 poses (ablation) | CamFormer · Ego-Exo4D text retrieval | top-1 accuracy (%): 43.86 -> 44.81 | No clear effect |
| finetune the text encoder instead of freezing (ablation) | CamFormer · Ego-Exo4D text retrieval | top-1 accuracy (%): 44.81 -> 45.42 | Improved |
| add CamFormer pretraining (ablation) | CamFormer · Ego-Exo4D proficiency (bouldering) | accuracy (%): 63.52 -> 65.41 | Improved |
| add CamFormer pretraining (ablation) | CamFormer · Ego-Exo4D activity classification (Aria poses) | accuracy (%): 61.83 -> 71.28 | Improved |
| add CamFormer pretraining (ablation) | CamFormer · UCF101-Dynamic (ViPE poses) | accuracy (%): 64.18 -> 68.16 | Improved |
| fuse trajectory features with video features | EgoVLPv2 · Ego-Exo4D keystep recognition | accuracy (%): 29.17 -> 32.37 | Improved |
Earlier methods this paper compares with
Named baselines from earlier accepted papers in this paper's results tables: better than 1 of 1 on most shared results.
- ShotVL-7B — ShotVL: Human-Centric Highlight Frame Retrieval via Language Queries (AAAI 2025) · better on 1 of 1 shared result
Matched by method name in results tables read by AI. Authors' own tables tend to favour the newer method, and a comparison can be missed when a method is renamed.
Closest papers by meaning
- Towards Understanding Camera Motions in Any Video (NeurIPS 2025 · similarity 0.87)
- Perception Encoder: The best visual embeddings are not at the output of the network (NeurIPS 2025 · similarity 0.87)
- Rays as Pixels: Learning A Joint Distribution of Video and Camera Trajectories (ICML 2026 · similarity 0.87)
- VividCam: Learning Unconventional Camera Motions from Virtual Synthetic Videos (ICML 2026 · similarity 0.86)
- Unveiling the Value of Motion for Cinematic Camera Trajectories (NeurIPS 2026 · similarity 0.85)
- Training-free Camera Control for Video Generation (ICLR 2025 · similarity 0.84)
Writing a paper in this area? scime draft is a free LaTeX editor with AI included, real-time co-authors and a check of your references and venue rules before you submit. It also imports OpenAI Prism projects.
