Research map / Architectures and efficiency / CNN design
LSTM-CF: Unifying Context Modeling and Fusion with LSTMs for RGB-D Scene Labeling
Zhen Li, Yukang Gan, Xiaodan Liang, Yizhou Yu, Hui Cheng, Liang Lin
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 the cross-layer connection of RGB conv features to the fused contexts (ablation) | LSTM-CF · SUNRGBD | mean accuracy (%): 15.2 -> 48.1 | Improved |
| add the RGB path (ablation) | LSTM-CF · SUNRGBD | mean accuracy (%): 15.8 -> 48.1 | Improved |
| add multi-scale RGB feature concatenation (ablation) | LSTM-CF · SUNRGBD | mean accuracy (%): 42.1 -> 48.1 | Improved |
| add the depth path (ablation) | LSTM-CF · SUNRGBD | mean accuracy (%): 43.7 -> 48.1 | Improved |
| add the memorized fusion layer (ablation) | LSTM-CF · SUNRGBD | mean accuracy (%): 44.7 -> 48.1 | Improved |
| add the memorized context layers (ablation) | LSTM-CF · SUNRGBD | mean accuracy (%): 45.7 -> 48.1 | Improved |
| add memorized layers (both context and fusion) versus none (ablation) | LSTM-CF · SUNRGBD | mean accuracy (%): 45.0 -> 48.1 | Improved |
| use separate memorized context layers per path instead of a single shared one before fusion (ablation) | LSTM-CF · SUNRGBD | mean accuracy | Improved |
| LSTM-CF versus previous state of the art | RGB-D scene labeling · SUNRGBD (37 classes) | average Jaccard index (%): 45.9 -> 48.1 | Improved |
| LSTM-CF versus previous state of the art | RGB-D scene labeling · NYUDv2 (37 classes) | average Jaccard index (%): 43.9 -> 49.4 | Improved |
| LSTM-CF versus previous state of the art | RGB-D scene labeling · SUN3D (12 classes) | average Jaccard index (%): 45.7 -> 58.5 | Improved |
| LSTM-CF versus previous method on test speed | RGB-D scene labeling · SUNRGBD | seconds per image: 1.5 -> 0.15 | Improved |
Closest papers by meaning
- Depth CNNs for RGB-D Scene Recognition: Learning from Scratch Better than Transferring from RGB-CNNs (AAAI 2017 · similarity 0.87)
- A Multi-Modal, Discriminative and Spatially Invariant CNN for RGB-D Object Labeling (TPAMI 2017 · similarity 0.86)
- SceneCode: Monocular Dense Semantic Reconstruction Using Learned Encoded Scene Representations (CVPR 2019 · similarity 0.86)
- Attention-Based Multi-Modal Fusion Network for Semantic Scene Completion (AAAI 2020 · similarity 0.85)
- Large Spatial Model: End-to-end Unposed Images to Semantic 3D (NeurIPS 2024 · similarity 0.85)
- Image Fusion via Vision-Language Model (ICML 2024 · similarity 0.84)
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.
