Research map / Detection and segmentation / CNN features
MCMAE: Masked Convolution Meets Masked Autoencoders
Peng Gao, Teli Ma, Hongsheng Li, Ziyi Lin, Jifeng Dai, Yu Qiao
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 masked convolution in stages 1-2 (ablation) | ConvMAE · ImageNet-1K finetuning | accuracy (%): 81.5 -> 84.6 | Improved |
| use block-wise masking instead of random input-token masking (ablation) | ConvMAE · ImageNet-1K finetuning | accuracy (%): 84.2 -> 84.6 | Improved |
| enlarge conv kernel from 5x5 to 7x7 / 9x9 in stages 1-2 (ablation) | ConvMAE · ImageNet-1K finetuning | accuracy (%): 84.5 -> 84.4 / 84.6 | No clear effect |
| add multi-scale decoder (200 epochs) (ablation) | ConvMAE-Base · ImageNet-1K finetuning | accuracy (%): 84.1 -> 84.4 | Improved |
| add multi-scale decoder (1600 epochs) (ablation) | ConvMAE-Base · ImageNet-1K finetuning | accuracy (%): 84.6 -> 85.0 | Improved |
| add multi-scale decoder (1600 epochs) (ablation) | ConvMAE-Base · COCO | AP box: 52.5 -> 53.2 | Improved |
| add multi-scale decoder (1600 epochs) (ablation) | ConvMAE-Base · ADE20K | mIoU: 50.7 -> 51.7 | Improved |
| increase pretraining epochs 200 -> 1600 (ablation) | ConvMAE-Base · ImageNet-1K linear probe | accuracy (%): 62.5 -> 69.4 | Improved |
| increase pretraining epochs 800 -> 1600 (ablation) | ConvMAE-Base · ImageNet-1K finetuning | accuracy (%): 84.6 -> 84.6 | No clear effect |
| replace ViT-B MAE with ConvMAE hybrid encoder | MAE · ImageNet-1K finetuning | accuracy (%): 83.6 -> 85.0 | Improved |
| replace ViT-B MAE with ConvMAE hybrid encoder | MAE · ADE20K | mIoU: 48.1 -> 51.7 | Improved |
Closest papers by meaning
- Unleashing Vanilla Vision Transformer with Masked Image Modeling for Object Detection (ICCV 2023 · similarity 0.89)
- OmniMAE: Single Model Masked Pretraining on Images and Videos (CVPR 2023 · similarity 0.89)
- Closest Neighbors are Harmful for Lightweight Masked Auto-encoders (CVPR 2025 · similarity 0.89)
- The Dynamic Duo of Collaborative Masking and Target for Advanced Masked Autoencoder Learning (AAAI 2025 · similarity 0.87)
- Asymmetric Masked Distillation for Pre-Training Small Foundation Models (CVPR 2024 · similarity 0.87)
- Learning Mask Invariant Mutual Information for Masked Image Modeling (ICLR 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.
