Research map / Detection and segmentation / CNN features
X-Learner: Learning Cross Sources and Tasks for Universal Visual Representation
Yinan He, Gengshi Huang, Siyu Chen, Jianing Teng, Kun Wang, Zhenfei Yin, Lu Sheng, Ziwei Liu, Yu Qiao, Jing Shao
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 reconciliation layers in expansion stage (ablation) | X-Learner · 10 classification datasets (linear probe) | average accuracy (%): 74.8 -> 77.1 | Improved |
| add reconciliation layers in expansion stage (ablation) | X-Learner · PASCAL VOC07 detection | mAP: 83.9 -> 84.4 | Improved |
| use X-Learner instead of hard-sharing multi-task | ResNet-50 · 10 classification datasets (linear probe) | average accuracy (%): 73.2 -> 77.1 | Improved |
| add extra pre-distillation hints (X-Learner++) | X-Learner · PASCAL VOC07 detection | mAP: 84.4 -> 84.8 | Improved |
| add semantic segmentation task and sources | X-Learner · PASCAL VOC12 segmentation | mIoU: 77.1 -> 77.6 | Improved |
| add semantic segmentation task and sources | X-Learner · PASCAL VOC07 detection | mAP: 84.4 -> 84.3 | Hurt |
| add classification sources (CompCars, Dogs) | Hard-sharing · ImageNet pre-training accuracy | accuracy (%): 75.0 -> 73.7 | Hurt |
| add classification sources (CompCars, Dogs) | X-Learner · 10 classification datasets (linear probe) | average accuracy (%): 77.1 -> 80.6 | Improved |
| reverse order to squeeze then expand (ablation) | X-Learner · 10 classification datasets (linear probe) | average accuracy (%): 77.1 -> 73.9 | Hurt |
| feed reconciliation layers from deeper layers instead of lower layers (ablation) | X-Learner · PASCAL VOC07 detection | mAP: 84.4 -> 83.5 | Hurt |
| replace distillation with pruning in squeeze stage (ablation) | X-Learner · 10 classification datasets (linear probe) | average accuracy (%): 77.1 -> 77.2 | No clear effect |
| replace distillation with pruning in squeeze stage (ablation) | X-Learner · PASCAL VOC07 detection | mAP: 84.4 -> 83.1 | Hurt |
Closest papers by meaning
- Universal Representations: A Unified Look at Multiple Task and Domain Learning (IJCV 2023 · similarity 0.88)
- Learning More Universal Representations for Transfer-Learning (TPAMI 2019 · similarity 0.87)
- Visual Bridge: Universal Visual Perception Representations Generating (AAAI 2026 · similarity 0.87)
- UniFormer: Unifying Convolution and Self-Attention for Visual Recognition (TPAMI 2023 · similarity 0.87)
- Perception Encoder: The best visual embeddings are not at the output of the network (NeurIPS 2025 · similarity 0.86)
- Sample- and Parameter-Efficient Auto-Regressive Image Models (CVPR 2025 · similarity 0.86)
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.
