Research map / Vision-language and multimodal / Human and embodied modeling
Broadcasting Convolutional Network for Visual Relational Reasoning
Simyung Chang, John Yang, SeongUk Park, Nojun Kwak
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 BCN (global max-pool broadcast) to the baseline CNN | 3-layer CNN · Scaled-MNIST | classification accuracy: 84.4% -> 97.5% | Improved |
| use max pooling instead of average pooling in BCN (ablation) | Base + BCN · Scaled-MNIST | classification accuracy: 92.7% -> 97.5% | Improved |
| add BCN on top of coordinates plus skip-connection (ablation) | Base + CCE + Skip-connection · Scaled-MNIST | localization error: 0.071 -> 0.023 | Improved |
| add coordinate channels (CCE) alone to baseline (ablation) | 3-layer CNN · Scaled-MNIST | localization error: 0.151 -> 0.088 | Improved |
| add BCN to baseline CNN | 4-layer CNN · STL-10 (from scratch) | accuracy: 68.75% -> 72.18% | Improved |
| use BCN instead of a plain skip-connection of the same convolutions (ablation) | 4-layer CNN · STL-10 (from scratch) | accuracy: 69.79% -> 72.18% | Improved |
| add BCN to ResNet-18 | ResNet-18 · STL-10 (from scratch) | accuracy: 76.27% -> 77.00% | Improved |
| add BCN and coordinate encoding to multiRN (ablation) | CNN+multiRN · Sort-of-CLEVR | relational accuracy: 88.7% -> 92.9% | Improved |
| add coordinate embedding to a CNN+MLP model (ablation) | CNN+MLP · Sort-of-CLEVR | relational accuracy: 74.2% -> 72.9% | Hurt |
| use stride 1 in the last conv layer (CNN_h) for multiRN | multiRN · Sort-of-CLEVR | relational accuracy: 92.9% -> 96.7% | Improved |
| replace RN with multiRN | CNN+LSTM · CLEVR (pixels) | overall accuracy: 90.9 -> 92.3 | Improved |
| use CNN_h with multiRN and enlarge input to 224x224 | CNN_h+LSTM+multiRN · CLEVR (pixels) | overall accuracy: 97.2 -> 97.7 | Improved |
Closest papers by meaning
- PARN: Position-Aware Relation Networks for Few-Shot Learning (ICCV 2019 · similarity 0.87)
- Detecting Visual Relationships With Deep Relational Networks (CVPR 2017 · similarity 0.87)
- A simple neural network module for relational reasoning (NeurIPS 2017 · similarity 0.87)
- Local Relation Networks for Image Recognition (ICCV 2019 · similarity 0.86)
- Graph-Based Global Reasoning Networks (CVPR 2019 · similarity 0.86)
- Composition-based Multi-Relational Graph Convolutional Networks (ICLR 2020 · similarity 0.86)
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.
