Research map / Video / Video segmentation and tracking
TRACK: A New Method from a Re-examination of Deep Architectures for Head Motion Prediction in 360-degree Videos
Miguel Fabián Romero Rondón, Lucile Sassatelli, Ramón Aparicio-Pardo, Fŕed́eric Precioso
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 |
|---|---|---|---|
| replace PAMI18 predictor with a trivial no-motion baseline | head motion prediction · PAMI18 dataset, H~30ms (1 frame) | Mean Overlap: 0.753 -> 0.968 | Improved |
| replace ChinaCom18 with the trivial-static baseline | head motion prediction · ChinaCom18 dataset, H=1s | accuracy: 72.54% -> 96.15% | Improved |
| replace NOSSDAV17-Orient. with the trivial-static baseline | head motion prediction · NOSSDAV17 dataset, H=1s | accuracy: 86.35% -> 95.79% | Improved |
| use seq2seq LSTM on positions instead of trivial-static | deep-position-only baseline · PAMI18 dataset, H~30ms (1 frame) | Mean Overlap: 0.968 -> 0.977 | Improved |
| use seq2seq LSTM on positions instead of trivial-static | deep-position-only baseline · ChinaCom18 dataset, H=1s | accuracy: 96.15% -> 96.90% | Improved |
| add video content (TRACK) to position-only baseline at short horizon | TRACK · NOSSDAV17 dataset, H=1s | accuracy: 96.30% -> 95.48% | Hurt |
| add seq2seq framework, 3D Euclidean MSE loss and displacement prediction to CVPR18 and MM18 (improved versions) (ablation) | CVPR18 and MM18 with ground-truth saliency · MMSys18 subset, H up to 5s | orthodromic distance | Improved |
| use a dedicated recurrent unit for positions before fusing with visual features (CVPR18 style) instead of one RNN doing both (MM18 style) (ablation) | head motion predictor with ground-truth saliency · MMSys18 subset, H up to 5s | orthodromic distance | Improved |
| replace ground-truth saliency with content-based saliency (ablation) | CVPR18-improved · MMSys18 subset, H up to 5s | orthodromic distance | Hurt |
| add a dedicated content RNN for saliency before fusion (TRACK) (ablation) | CVPR18-improved with content-based saliency · MMSys18 subset, H up to 5s | orthodromic distance | Improved |
Later papers that compared with this method
1 later accepted paper reports Track as a baseline in their results tables; 1 of them report a better number on most of the results they share with it.
- Learning To Associate Every Segment for Video Panoptic Segmentation (CVPR 2021 · better on 3 of 3 shared results)
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
- A Neural Temporal Model for Human Motion Prediction (CVPR 2019 · similarity 0.88)
- Tracktention: Leveraging Point Tracking to Attend Videos Faster and Better (CVPR 2025 · similarity 0.88)
- Leader360V: A Large-scale, Real-world 360 Video Dataset for Multi-task Learning in Diverse Environment (NeurIPS 2025 · similarity 0.85)
- Track-On: Transformer-based Online Point Tracking with Memory (ICLR 2025 · similarity 0.85)
- TAPVid-360: Tracking Any Point in 360 from Narrow Field of View Video (NeurIPS 2025 · similarity 0.85)
- Motion Graph Unleashed: A Novel Approach to Video Prediction (NeurIPS 2024 · similarity 0.85)
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.
