Research map / Label-efficient and robust learning / Domain adaptation and generalization
To Adapt or Not to Adapt? Real-Time Adaptation for Semantic Segmentation
Marc Botet Colomer, Pier Luigi Dovesi, Theodoros Panagiotakopoulos, Joao Frederico Carvalho, Linus Härenstam-Nielsen, Hossein Azizpour, Hedvig Kjellström, Daniel Cremers, Matteo Poggi
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 HAMT (hardware-aware backprop orchestration) to full online training (ablation) | HAMLET · Increasing Storm | FPS / all-domain mIoU: 5.9 -> 7.0 FPS; 69.5 -> 68.7 mIoU | Mixed |
| add Least Training (adapt only on detected domain shift) (ablation) | HAMLET with HAMT · Increasing Storm | FPS / 200mm mIoU: 7.0 -> 29.5 FPS; 60.2 -> 51.8 mIoU (200mm) | Mixed |
| add Adaptive Learning Rate to LT (ablation) | HAMLET · Increasing Storm | mIoU (200mm): 51.8 -> 54.1 | Improved |
| add domain-conditioned module and random class sampling on top of ALR (ablation) | HAMLET · Increasing Storm | mIoU (200mm): 54.1 -> 58.2 | Improved |
| use Least Training alone when domain shifts are frequent (ablation) | HAMLET · Fast Storm C | FPS: 5.6 -> 7.2 FPS (A vs C) | No clear effect |
| full HAMLET with all components vs full online training (ablation) | HAMLET · Fast Storm C | total h-mean mIoU: 67.2 -> 67.8 | Improved |
| run CoTTA in real-time (adapt once every 50 frames) | CoTTA · OnDA Increasing Storm | h-mean mIoU (total): 66.8 -> 62.3 | Hurt |
| use HAMLET instead of TENT for online adaptation | SegFormer MiT-B1 · OnDA Increasing Storm | h-mean mIoU (total): 59.9 -> 67.2 | Improved |
| adapt SegFormer MiT-B1 with HAMLET vs source model only | SegFormer MiT-B1 · OnDA Increasing Storm | h-mean mIoU (forward): 57.8 -> 66.8 | Improved |
| use a TENT replay buffer | TENT · OnDA Increasing Storm | h-mean mIoU (total): 59.9 -> 59.9 | No clear effect |
Earlier methods this paper compares with
Named baselines from earlier accepted papers in this paper's results tables: better than 1 of 1 on most shared results.
- TENT — Tent: Fully Test-Time Adaptation by Entropy Minimization (ICLR 2021) · better on 1 of 1 shared result
Matched by method name in results tables read by AI. Authors' own tables tend to favour the newer method, and a comparison can be missed when a method is renamed.
Closest papers by meaning
- On the Road to Online Adaptation for Semantic Image Segmentation (CVPR 2022 · similarity 0.92)
- Online Domain Adaptation for Semantic Segmentation in Ever-Changing Conditions (ECCV 2022 · similarity 0.91)
- D2ADA: Dynamic Density-Aware Active Domain Adaptation for Semantic Segmentation (ECCV 2022 · similarity 0.90)
- Multi-source Domain Adaptation for Semantic Segmentation (NeurIPS 2019 · similarity 0.89)
- Source-Free Domain Adaptation for Semantic Segmentation (CVPR 2021 · similarity 0.89)
- Learning to Adapt via Latent Domains for Adaptive Semantic Segmentation (NeurIPS 2021 · similarity 0.89)
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.
