Research map / Label-efficient and robust learning / Domain adaptation and generalization
Learning Good Features to Transfer Across Tasks and Domains
Pierluigi Zama Ramirez, Adriano Cardace, Luca De Luigi, Alessio Tonioni, Samuele Salti, Luigi Di Stefano
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 edge detection auxiliary task (ablation) | base AT/DT · Carla to Cityscapes, depth to semantics | mIoU (%): 48.04 -> 50.16 | Improved |
| add NDA loss (ablation) | base AT/DT · Carla to Cityscapes, depth to semantics | mIoU (%): 48.04 -> 50.30 | Improved |
| add both edge auxiliary task and NDA loss (ablation) | base AT/DT · Carla to Cityscapes, depth to semantics | mIoU (%): 48.04 -> 51.28 | Improved |
| use autoencoder instead of edge detection as auxiliary task (ablation) | base AT/DT · Carla to Cityscapes, depth to semantics | mIoU (%): 48.04 -> 48.07 | No clear effect |
| use edge detection instead of depth as source task (ablation) | base AT/DT · Carla to Cityscapes, depth to semantics | mIoU (%): 48.04 -> 20.38 | Hurt |
| add edge decoder while training N1 only (ablation) | base AT/DT · Carla to Cityscapes, depth to semantics | mIoU (%): 48.04 -> 48.69 | Improved |
| use two separate edge decoders for N1 and N2 (ablation) | base AT/DT · Carla to Cityscapes, depth to semantics | mIoU (%): 48.04 -> 49.99 | Improved |
| train N1, N2 and a shared edge decoder simultaneously (ablation) | separate edge decoders · Carla to Cityscapes, depth to semantics | mIoU (%): 49.99 -> 50.16 | Improved |
| add adversarial alignment of E1 features (ablation) | base AT/DT · Carla to Cityscapes, depth to semantics | mIoU (%): 48.04 -> 48.19 | No clear effect |
| add asymmetric adversarial alignment of E1 features (ablation) | base AT/DT · Carla to Cityscapes, depth to semantics | mIoU (%): 48.04 -> 44.00 | Hurt |
| add LargerNorm with norm 500 (ablation) | base AT/DT · Carla to Cityscapes, depth to semantics | mIoU (%): 48.04 -> 35.08 | Hurt |
| add NDA loss on E2 features (ablation) | base AT/DT · Carla to Cityscapes, depth to semantics | mIoU (%): 48.04 -> 27.18 | Hurt |
Closest papers by meaning
- Learning Across Tasks and Domains (ICCV 2019 · similarity 0.92)
- Learning To Relate Depth and Semantics for Unsupervised Domain Adaptation (CVPR 2021 · similarity 0.90)
- Transferable Representation Learning with Deep Adaptation Networks (TPAMI 2018 · similarity 0.88)
- Deep Supervision with Intermediate Concepts (TPAMI 2018 · similarity 0.88)
- Cross-Domain Self-Supervised Multi-Task Feature Learning Using Synthetic Imagery (CVPR 2018 · similarity 0.88)
- DESC: Domain Adaptation for Depth Estimation via Semantic Consistency (IJCV 2022 · similarity 0.88)
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.
