Research map / Detection and segmentation / CNN features
MAMBA: Multi-level Aggregation via Memory Bank for Video Object Detection
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 |
|---|---|---|---|
| Add MAMBA memory-bank enhancement | RFCN with ResNet-101 · ImageNet VID validation | mAP: 73.8 -> 80.8 | Helped |
| Add MAMBA memory-bank enhancement | FasterRCNN with ResNet-101 · ImageNet VID validation | mAP: 75.4 -> 84.6 | Helped |
| Add random crop and random scale training augmentations | MAMBA with FasterRCNN and ResNeXt-101 · ImageNet VID validation | mAP: 85.4 -> 86.7 | Helped |
| Add instance-level memory enhancement (ablation) | RFCN · ImageNet VID validation | mAP: 73.8 -> 76.7 | Helped |
| Add instance-level memory enhancement (ablation) | FasterRCNN · ImageNet VID validation | mAP: 75.4 -> 83.7 | Helped |
| Add pixel-level memory enhancement (ablation) | FasterRCNN · ImageNet VID validation | mAP: 75.4 -> 81.8 | Helped |
| Add feature-wise instead of frame-wise memory updating (ablation) | SELSA baseline · ImageNet VID validation | mAP: 81.7 -> 82.4 | Helped |
| Add class-wise memory maintained across videos (ablation) | Feature-wise updating model · ImageNet VID validation | mAP: 82.4 -> 82.7 | Helped |
| Sample a subset of memory features instead of concatenating all features (ablation) | Model (d), N_m=24k · ImageNet VID validation | mAP: 82.7 -> 83.7 | Helped |
| Sample a subset of memory features instead of concatenating all features (ablation) | Model (d), N_m=24k · ImageNet VID validation | runtime: 142.9 -> 79.6 | Helped |
| Increase memory bank size while using sampled keys (ablation) | MAMBA · ImageNet VID validation, N_m=24k to 96k | mAP: 83.7 -> 83.8 | Helped |
| Reduce sampled key set from N_k=5k to N_k=50 (ablation) | MAMBA instance-level-only model · ImageNet VID validation | mAP: 83.8 -> 83.1 | Hurt |
Closest papers by meaning
- Mamba YOLO: A Simple Baseline for Object Detection with State Space Model (AAAI 2025 · similarity 0.87)
- MambaOut: Do We Really Need Mamba for Vision? (CVPR 2025 · similarity 0.87)
- Samba: A Unified Mamba-based Framework for General Salient Object Detection (CVPR 2025 · similarity 0.86)
- SMamba: Sparse Mamba for Event-based Object Detection (AAAI 2025 · similarity 0.86)
- Vision Mamba Mender (NeurIPS 2024 · similarity 0.85)
- Mamba-Reg: Vision Mamba Also Needs Registers (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.