Research map / 3D vision / Human pose estimation
A Dual-Transformer Architecture with Cross-Attention for Multi-Camera View Recommendation
Josep Cabacas Maso, Carles Ventura, Ismael Benito-Altamirano
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 |
|---|---|---|---|
| add focal loss in place of BCE (ablation) | Dual-Transformer (SwinV1) · TVMCE | Precision@0.5 (%): 52.60 -> 56.60 | Improved |
| decouple history and candidates with dual-transformer cross-attention vs single transformer (BCE + SwinV1) | Single Transformer re-impl. · TVMCE | Precision@0.5 (%): 47.97 -> 52.60 | Improved |
| decouple history and candidates vs single transformer (BCE + SwinV2) | Single Transformer re-impl. · TVMCE | Precision@0.5 (%): 54.06 -> 69.65 | Improved |
| replace SwinV1 backbone with SwinV2 (ablation) | Dual-Transformer (Focal) · TVMCE | Precision@0.5 (%): 56.60 -> 69.65 | Improved |
| replace SwinV2 with MaxViT-Tiny backbone (ablation) | Dual-Transformer (Focal) · TVMCE | Precision@0.5 (%): 69.65 -> 58.63 | Hurt |
| replace SwinV2 with ViT-Base backbone (ablation) | Dual-Transformer (Focal) · TVMCE | Precision@0.5 (%): 69.65 -> 25.85 | Hurt |
| replace SwinV2 with ConvNeXtV2-Tiny backbone (ablation) | Dual-Transformer (Focal) · TVMCE | Precision@0.5 (%): 69.65 -> 56.23 | Hurt |
| replace SwinV2 with ResNet-50 backbone (ablation) | Dual-Transformer (Focal) · TVMCE | Precision@0.5 (%): 69.65 -> 46.74 | Hurt |
| lower decision threshold from 0.5 to 0.3 | Dual-Transformer (SwinV2, Focal) · TVMCE validation (macro) | F1 (%): 74.53 -> 81.41 | Improved |
| fine-tune on 20% of target video vs zero-shot | Dual-Transformer · video_0000 | Precision@0.5 (%): 74.44 -> 83.65 | Improved |
| fine-tune on 30% of target video vs zero-shot | Dual-Transformer · video_0000 | Precision@0.5 (%): 74.44 -> 90.41 | Improved |
| fine-tune on 10% of video_0001 vs zero-shot | Dual-Transformer · video_0001 | Precision@0.5 (%): 81.76 -> 77.06 | Hurt |
Closest papers by meaning
- AHOY! Animatable Humans under Occlusion from YouTube Videos with Gaussian Splatting and Video Diffusion Priors (ECCV 2026 · similarity 0.89)
- End-to-End Multi-Person Pose Estimation with Pose-Aware Video Transformer (AAAI 2026 · similarity 0.85)
- GAF: Gaussian Avatar Reconstruction from Monocular Videos via Multi-view Diffusion (CVPR 2025 · similarity 0.84)
- One View Is Enough: In-the-Wild Monocular Pretraining for Novel View Generation (NeurIPS 2026 · similarity 0.84)
- 3D²-Actor: Learning Pose-Conditioned 3D-Aware Denoiser for Realistic Gaussian Avatar Modeling (AAAI 2025 · similarity 0.84)
- Sharingan: A Transformer Architecture for Multi-Person Gaze Following (CVPR 2024 · similarity 0.83)
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.
