Research map / Architectures and efficiency / Attention and transformers
TESA: Tensor Element Self-Attention via Matricization
Francesca Babiloni, Ioannis Marras, Gregory Slabaugh, Stefanos Zafeiriou
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 TESA at layers 2, 4, and 6 | PConv · CelebA inpainting | PSNR: 25.66 / 25.57 -> 26.00 / 25.81 | Improved |
| Add TESA at layers 2, 4, and 6 | PConv · CelebA inpainting | SSIM: 0.888 / 0.885 -> 0.895 / 0.891 | Improved |
| Add TESA at layers 2, 4, and 6 | PConv · CelebA inpainting | MS-SSIM: 0.931 / 0.927 -> 0.936 / 0.931 | Improved |
| Add TESA between the encoder and decoder | SID · SID Sony raw-to-rgb | PSNR: 29.62 -> 29.79 | Improved |
| Add TESA between the encoder and decoder | SID · SID Sony raw-to-rgb | SSIM: 0.889 -> 0.891 | Improved |
| Add one TESA block | ResNet50 · ImageNet classification | Top 1: 77.69 -> 76.49 | Hurt |
| Add one TESA block | ResNet50 · ImageNet classification | Top 5: 93.64 -> 93.05 | Hurt |
| Add one TESA block | Mask R-CNN · MS-COCO instance segmentation | APbox: 39.1 -> 39.5 | Improved |
| Add one TESA block | Mask R-CNN · MS-COCO instance segmentation | APbox50: 61.1 -> 61.2 | Improved |
| Add one TESA block | Mask R-CNN · MS-COCO instance segmentation | APbox75: 42.5 -> 43.0 | Improved |
| Add one TESA block | Mask R-CNN · MS-COCO instance segmentation | APmask: 35.5 -> 35.7 | Improved |
| Add one TESA block | Mask R-CNN · MS-COCO instance segmentation | APmask50: 58.0 -> 57.9 | Hurt |
Closest papers by meaning
- Attention-based Iterative Decomposition for Tensor Product Representation (ICLR 2024 · similarity 0.85)
- Convolutions and More as Einsum: A Tensor Network Perspective with Advances for Second-Order Methods (NeurIPS 2024 · similarity 0.84)
- Coarse-To-Fine Tensor Trains for Compact Visual Representations (ICML 2024 · similarity 0.84)
- Fully Distributed, Flexible Compositional Visual Representations via Soft Tensor Products (NeurIPS 2024 · similarity 0.84)
- Unveiling the Hidden Structure of Self-Attention via Kernel Principal Component Analysis (NeurIPS 2024 · similarity 0.84)
- Repurposing Stable Diffusion Attention for Training-Free Unsupervised Interactive Segmentation (CVPR 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.
