Research map / 3D vision / Depth and stereo
MAMo: Leveraging Memory and Attention for Monocular Video Depth Estimation
Rajeev Yasarla, Hong Cai, Jisoo Jeong, Yunxiao Shi, Risheek Garrepalli, Fatih Porikli
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 MAMo memory and attention to monocular depth network | PixelFormer Swin-large · KITTI Eigen | RMSE: 2.093 -> 1.884 | Improved |
| add MAMo to NeWCRFs for zero-shot transfer | NeWCRFs Swin-large · DDAD (KITTI-trained) | Sq Rel: 4.041 -> 2.990 | Improved |
| add MAMo to ResNet-DPT | ResNet-DPT ResNet50 · NYUv2-Video | RMSE: 0.464 -> 0.420 | Improved |
| carry over previous decoder features Ft-1 (ablation) | NeWCRFs Swin-Base · NYUv2-Video | RMSE: 0.3838 -> 0.3782 (1.5% decrease) | Improved |
| warp previous decoder features with optical flow before decoder (ablation) | NeWCRFs Swin-Base · NYUv2-Video | RMSE: 0.3782 -> 0.3779 | No clear effect |
| concatenate Ot and Ft-1 as decoder inputs (ablation) | NeWCRFs Swin-Base · NYUv2-Video | RMSE: 0.3838 -> 0.3743 (2.5% decrease) | Improved |
| sliding-window memory instead of proposed memory update (ablation) | NeWCRFs Swin-Base · NYUv2-Video | RMSE: 0.3838 -> 0.3827 | No clear effect |
| add proposed memory update scheme (ablation) | NeWCRFs Swin-Base · NYUv2-Video | RMSE: 0.3838 -> 0.3722 (3.5% decrease) | Improved |
| use full MAMo instead of plain NeWCRFs (ablation) | NeWCRFs Swin-Base · NYUv2-Video | RMSE: 0.3838 -> 0.3708 | Improved |
| use RAFT-small instead of RAFT for optical flow (ablation) | MAMo · NYUv2-Video | RMSE: 0.353 -> 0.355 | No clear effect |
| increase memory length L from 2 to 4 (ablation) | MAMo · KITTI | RMSE: 2.030 -> 2.003 | Improved |
| increase memory length L from 4 to 6 (ablation) | MAMo · KITTI | RMSE: 2.003 -> 1.990 | No clear effect |
Closest papers by meaning
- Self-Supervised Monocular Trained Depth Estimation Using Self-Attention and Discrete Disparity Volume (CVPR 2020 · similarity 0.89)
- Video Depth Anything: Consistent Depth Estimation for Super-Long Videos (CVPR 2025 · similarity 0.88)
- Multi-view Depth Estimation using Epipolar Spatio-Temporal Networks (CVPR 2021 · similarity 0.88)
- Exploiting Temporal Consistency for Real-Time Video Depth Estimation (ICCV 2019 · similarity 0.88)
- Trap Attention: Monocular Depth Estimation With Manual Traps (CVPR 2023 · similarity 0.88)
- Buffer Anytime: Zero-Shot Video Depth and Normal from Image Priors (CVPR 2025 · similarity 0.87)
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.
