Research map / Detection and segmentation / CNN features
Spatial Memory for Context Reasoning in Object Detection
Xinlei Chen, Abhinav Gupta
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 flipped-label region sampling (ablation) | SMN Base · VOC 2007 test (N=5) | mAP: 63.6 -> 64.4 | Improved |
| add reconstruction loss on top of flipped sampling (ablation) | SMN · VOC 2007 test (N=5) | mAP: 64.4 -> 64.6 | Improved |
| add flipped sampling and reconstruction loss (ablation) | SMN Base · COCO 2014 minival (N=5) | AP: 23.3 -> 23.8 | Improved |
| tune N=10 model from the N=5 model (ablation) | SMN Full · COCO 2014 minival (N=10) | AP: 26.6 -> 27.1 | Improved |
| use SoftMax instead of HardMax class assignment (ablation) | SMN Full · VOC 2007 test (N=5) | mAP: 64.6 -> 66.4 | Improved |
| use SoftMax instead of HardMax class assignment (ablation) | SMN Full · VOC 2007 test (N=10) | mAP: 67.8 -> 69.4 | Improved |
| use HardMax instead of SoftMax for baseline NMS (ablation) | Baseline FRCNN · VOC 2007 test (N=10) | mAP: 70.3 -> 68.0 | Hurt |
| use non-aggressive NMS (top-5k RoIs) in SMN (ablation) | SMN Full · COCO 2014 minival (N=10) | AP: 27.7 -> 28.1 | Improved |
| use non-aggressive NMS (top-5k RoIs) in SMN (ablation) | SMN Full · VOC 2007 test (N=10) | mAP: 69.4 -> 70.0 | Mixed |
| replace spatial memory with a 5-layer ConvNet MLP on conv5_3 | SMN · COCO 2014 minival | AP: 31.6 -> 30.1 | Hurt |
| add SMN context reasoning to Faster R-CNN | Baseline FRCNN · COCO 2014 minival | AP: 29.4 -> 31.6 | Improved |
| add SMN context reasoning to Faster R-CNN | Baseline FRCNN · VOC 2007 test (final) | mAP: 71.2 -> 71.1 | No clear effect |
Closest papers by meaning
- Reasoning-based Spatial Prior (RSP): Learning Spatial Priors from Multimodal LLMs for Object Detection (NeurIPS 2026 · similarity 0.89)
- Spatial-Aware Graph Relation Network for Large-Scale Object Detection (CVPR 2019 · similarity 0.88)
- Spatially Invariant Unsupervised Object Detection with Convolutional Neural Networks (AAAI 2019 · similarity 0.88)
- Structure Inference Net: Object Detection Using Scene-Level Context and Instance-Level Relationships (CVPR 2018 · similarity 0.88)
- Why Is Spatial Reasoning Hard for VLMs? An Attention Mechanism Perspective on Focus Areas (ICML 2025 · similarity 0.86)
- ContextShift: A Controlled Benchmark for Context Dependence in Object Detection (NeurIPS 2026 · 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.
