Research map / Continual learning / Class-incremental
RECALL: Replay-Based Continual Learning in Semantic Segmentation
Andrea Maracani, Umberto Michieli, Marco Toldo, Pietro Zanuttigh
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 Web replay data on top of background inpainting (ablation) | RECALL · Pascal VOC2012 10-1 disjoint | mIoU: 34.0 -> 57.8 | Improved |
| add GAN replay data on top of background inpainting (ablation) | RECALL · Pascal VOC2012 10-1 disjoint | mIoU: 34.0 -> 53.9 | Improved |
| add background inpainting on top of Web replay (ablation) | RECALL · Pascal VOC2012 10-1 disjoint | mIoU: 45.4 -> 57.8 | Improved |
| add background inpainting on top of GAN replay (ablation) | RECALL · Pascal VOC2012 15-5 disjoint | mIoU: 49.8 -> 63.5 | Improved |
| add background inpainting on top of Web replay (ablation) | RECALL · Pascal VOC2012 10-10 disjoint | mIoU: 55.2 -> 61.9 | Improved |
| use Web replay data instead of GAN replay data | RECALL · Pascal VOC2012 10-1 disjoint | mIoU: 53.9 -> 57.8 | Improved |
| replace stored real samples (S&R) with RECALL Web replay | S&R baseline · Pascal VOC2012 15-1 disjoint | mIoU: 40.7 -> 64.3 | Improved |
| initialize backbone with MS COCO instead of ImageNet | RECALL (Web) · Pascal VOC2012 10-1 disjoint | mIoU: 57.8 -> 64.3 | Improved |
| initialize backbone with MS COCO instead of ImageNet | RECALL (GAN) · Pascal VOC2012 10-10 disjoint | mIoU: 60.8 -> 67.6 | Improved |
| add RECALL replay (ours, Flickr) to MiB | MiB · Pascal VOC2012 15-1 disjoint | mIoU: 37.9 -> 65.7 | Improved |
| add RECALL replay (ours, GAN) to SDR | SDR · Pascal VOC2012 15-1 disjoint | mIoU: 48.1 -> 65.5 | Improved |
| add RECALL replay (ours, Flickr) to ILT | ILT · Pascal VOC2012 15-1 disjoint | mIoU: 5.4 -> 51.5 | Improved |
Closest papers by meaning
- Replay Master: Automatic Sample Selection and Effective Memory Utilization for Continual Semantic Segmentation (TPAMI 2025 · similarity 0.92)
- Beyond Background Shift: Rethinking Instance Replay in Continual Semantic Segmentation (CVPR 2025 · similarity 0.91)
- Adaptive Prototype Replay for Class Incremental Semantic Segmentation (AAAI 2025 · similarity 0.88)
- AAKR: Adversarial Attack-based Knowledge Retention for Continual Semantic Segmentation (AAAI 2025 · similarity 0.88)
- Towards Continual Universal Segmentation (CVPR 2025 · similarity 0.87)
- Enhancing Visual Continual Learning with Language-Guided Supervision (CVPR 2024 · similarity 0.87)
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.
