Research map / Detection and segmentation / CNN features
Deep Set Prediction Networks
Read the paperCheck your idea against 108,967 papers
AI-extracted summary. Written by AI from the paper's title only (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 |
|---|---|---|---|
| decode sets by gradient descent through a set encoder instead of MLP decoder | set auto-encoder · MNIST | Chamfer loss: 0.21 -> 0.09 | Helped |
| decode sets by gradient descent instead of RNN decoder | set auto-encoder · MNIST | Chamfer loss: 0.49 -> 0.09 | Helped |
| use DSPN instead of MLP decoder | ResNet34 image encoder · CLEVR bounding boxes | AP95: 57.9 -> 85.7 | Helped |
| use DSPN instead of MLP decoder at lenient IoU | ResNet34 image encoder · CLEVR bounding boxes | AP50: 99.3 -> 98.8 | No clear effect |
| use DSPN instead of MLP decoder | ResNet34 image encoder · CLEVR state prediction | AP_inf: 3.6 -> 72.8 | Helped |
| double inner optimisation steps from 10 to 20 at test time (ablation) | DSPN · CLEVR bounding boxes | AP98: 34.5 -> 24.3; AP90 94.3 -> 98.7 | Mixed |
| double inner optimisation steps from 10 to 20 at test time (ablation) | DSPN · CLEVR state prediction | AP0.5: 39.0 -> 57.0 | Helped |
| increase inner steps from 20 to 30 at test time (ablation) | DSPN · CLEVR state prediction | AP0.5: 57.0 -> 47.4 | Hurt |
| add representation loss term on target set (lambda) to outer loss (ablation) | DSPN · supervised set prediction | training stability and convergence | Helped |
| compute set loss over all intermediate sets instead of only the last (ablation) | DSPN · CLEVR state prediction | AP0.5: 57.0 -> 66.8 | Helped |
| compute set loss over all intermediate sets instead of only the last (ablation) | DSPN · CLEVR bounding boxes | AP90: 16.0 -> 3.3 | Hurt |
Closest papers by meaning
- Joint Learning of Set Cardinality and State Distribution (AAAI 2018 · similarity 0.87)
- Multiset-Equivariant Set Prediction with Approximate Implicit Differentiation (ICLR 2022 · similarity 0.87)
- Set Prediction in the Latent Space (NeurIPS 2021 · similarity 0.85)
- Polynomial Width is Sufficient for Set Representation with High-dimensional Features (ICLR 2024 · similarity 0.81)
- Set-based Neural Network Encoding Without Weight Tying (NeurIPS 2024 · similarity 0.81)
- Unlocking Better Closed-Set Alignment Based on Neural Collapse for Open-Set Recognition (AAAI 2025 · similarity 0.80)
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.