Research map / Label-efficient and robust learning / Recognition and long-tail
Cross-Stitch Networks for Multi-Task Learning
Check your idea against 108,967 papers
AI-extracted summary. Written by AI from the paper's title only (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 cross-stitch sharing instead of the strongest reported baseline | AlexNet for surface normal prediction · NYU-v2 | Mean angular error: 34.4 -> 34.1 | Helped |
| Use cross-stitch sharing instead of the strongest reported split baseline | AlexNet for semantic segmentation · NYU-v2 | mIU: 19.2 -> 19.3 | Helped |
| Use cross-stitch sharing instead of the strongest reported baseline | Fast R-CNN AlexNet for object detection · PASCAL VOC 2008 | Detection mAP: 46.1 -> 45.2 | Hurt |
| Use cross-stitch sharing instead of the strongest reported baseline | Fast R-CNN AlexNet for attribute prediction · PASCAL VOC 2008 | Attributes mAP: 61.1 -> 63.0 | Helped |
| Initialize αS=0.9 and αD=0.1 rather than αS=0.1 and αD=0.9 (ablation) | Cross-stitch network · NYU-v2 | Mean angular error: 34.6 -> 34.0 | Helped |
| Increase cross-stitch unit learning-rate scale from 1 to 10² (ablation) | Cross-stitch network · NYU-v2 | Mean angular error: 34.6 -> 34.0 | Helped |
| Increase cross-stitch unit learning-rate scale from 1 to 10³ (ablation) | Cross-stitch network · NYU-v2 | Mean angular error: 34.6 -> 34.1 | Helped |
| Initialize networks from task-specific one-task models rather than ImageNet (ablation) | Cross-stitch network · NYU-v2 | Mean angular error: 34.6 -> 34.1 | Helped |
| Use one cross-stitch unit per activation map rather than one per channel (ablation) | Object detection and attribute prediction network · PASCAL VOC 2008 | Detection mAP | Helped |
Closest papers by meaning
- End-To-End Multi-Task Learning With Attention (CVPR 2019 · similarity 0.87)
- AdaShare: Learning What To Share For Efficient Deep Multi-Task Learning (NeurIPS 2020 · similarity 0.87)
- Which Tasks Should Be Learned Together in Multi-task Learning? (ICML 2020 · similarity 0.87)
- How Not to Stitch Representations to Measure Similarity: Task Loss Matching Versus Direct Matching (AAAI 2025 · similarity 0.86)
- Provable Multi-Task Representation Learning by Two-Layer ReLU Neural Networks (ICML 2024 · similarity 0.85)
- Going Beyond Multi-Task Dense Prediction with Synergy Embedding Models (CVPR 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.