Research map / Image generation / Image restoration
CompositeTasking: Understanding Images by Spatial Composition of Tasks
Nikola Popovic, Danda Pani Paudel, Thomas Probst, Guolei Sun, Luc Van Gool
Check 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 |
|---|---|---|---|
| Use spatial task conditioning with one shared encoder-decoder, trained on random mosaic label regions | CTN · PASCAL-MT, single-task rule S | mIoU: 59.82 -> 62.45 | Improved |
| Use spatial task conditioning with one shared encoder-decoder, trained on random mosaic label regions | CTN · PASCAL-MT, single-task rule S | mErr: 16.07 -> 16.93 | Hurt |
| Use spatial task conditioning for sparse, semantically assigned task regions | CTN · PASCAL-MT, semantic rule R2 | odsF: 63.90 -> 69.20 | Improved |
| Use spatial task conditioning for sparse, semantically assigned task regions | CTN · PASCAL-MT, semantic rule R2 | mErr: 17.13 -> 18.12 | Hurt |
| Fine-tune the R2-trained model on the similar R3 rule | CTN · PASCAL-MT, semantic rule R3 | mErr: 20.11 -> 18.68 | Improved |
| Fine-tune the R2-trained model on R3 | CTN · PASCAL-MT, semantic rule R2 | mErr: 18.12 -> 17.95 | Improved |
| Train on independently random task assignments instead of large random mosaics | CTN · PASCAL-MT, random task per pixel Rrnd | mIoU: 35.89 -> 59.58 | Improved |
| Train on independently random task assignments instead of large random mosaics | CTN · PASCAL-MT, random task per pixel Rrnd | mErr: 64.36 -> 17.16 | Improved |
| Use 2 shared embedding convolutions instead of 1 (ablation) | CTN · PASCAL-MT, single-task rule S | mIoU: 62.45 -> 61.93 | Hurt |
| Replace 3 × 3 decoder convolution with 1 × 1 (ablation) | CTN · PASCAL-MT, single-task rule S | mIoU: 62.45 -> 61.50 | Hurt |
| Use 6 output channels instead of 3 (ablation) | CTN · PASCAL-MT, single-task rule S | mIoU: 62.45 -> 62.35 | No clear effect |
| Replace bilinear with nearest-neighbor palette-embedding interpolation (ablation) | CTN · PASCAL-MT, single-task rule S | mIoU: 62.45 -> 60.92 | Hurt |
Closest papers by meaning
- TaskPrompter: Spatial-Channel Multi-Task Prompting for Dense Scene Understanding (ICLR 2023 · similarity 0.88)
- DiSparse: Disentangled Sparsification for Multitask Model Compression (CVPR 2022 · similarity 0.86)
- Clustered Task-Aware Meta-Learning by Learning From Learning Paths (TPAMI 2023 · similarity 0.86)
- Going Beyond Multi-Task Dense Prediction with Synergy Embedding Models (CVPR 2024 · similarity 0.86)
- Amalgamating Multi-Task Models with Heterogeneous Architectures (AAAI 2024 · similarity 0.85)
- Task-driven Image Fusion with Learnable Fusion Loss (CVPR 2025 · 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.
