Research map / Label-efficient and robust learning / Self-supervised and contrastive
Self-Supervised Visual Representations Learning by Contrastive Mask Prediction
Yucheng Zhao, Guangting Wang, Chong Luo, Wenjun Zeng, Zheng-Jun Zha
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 mask prediction head (3 blocks vs none) (ablation) | MaskCo · ImageNet linear conv5 (pretrained on ImageNet) | top-1 accuracy: 41.9 -> 63.1 | Improved |
| add mask prediction head (3 blocks vs none) (ablation) | MaskCo · Pascal VOC detection (pretrained on ImageNet) | AP: 56.1 -> 56.0 | No clear effect |
| increase MPH from 1 to 3 blocks (ablation) | MaskCo · ImageNet linear conv5 (pretrained on ImageNet) | top-1 accuracy: 59.0 -> 63.1 | Improved |
| increase MPH from 3 to 4 blocks (ablation) | MaskCo · ImageNet linear conv5 (pretrained on ImageNet) | top-1 accuracy: 63.1 -> 63.0 | No clear effect |
| add masking of the query region (ablation) | MaskCo without MPH · Pascal VOC detection (pretrained on ImageNet) | AP: 53.6 -> 56.1 | Improved |
| add masking of the query region (ablation) | MaskCo without MPH · ImageNet linear conv4 (pretrained on ImageNet) | top-1 accuracy: 51.9 -> 57.2 | Improved |
| add masking of the query region (ablation) | MaskCo without MPH · ImageNet linear conv5 (pretrained on ImageNet) | top-1 accuracy: 52.3 -> 41.9 | Hurt |
| add masking of the query region (ablation) | MaskCo without MPH · Pascal VOC detection (pretrained on COCO) | AP: 54.1 -> 56.2 | Improved |
| add intra-image negatives to inter-image negatives (ablation) | MaskCo · Pascal VOC detection (pretrained on ImageNet) | AP: 56.0 -> 56.4 | Improved |
| add intra-image negatives to inter-image negatives (ablation) | MaskCo · ImageNet linear conv5 (pretrained on ImageNet) | top-1 accuracy: 63.1 -> 61.3 | Hurt |
| use only intra-image negatives instead of inter-image (ablation) | MaskCo · ImageNet linear conv5 (pretrained on ImageNet) | top-1 accuracy: 63.1 -> 48.6 | Hurt |
| replace MoCo v2 with MaskCo as pretraining method on COCO | COCO pretraining · ImageNet linear conv5 (pretrained on COCO) | top-1 accuracy: 51.4 -> 56.7 | Improved |
Closest papers by meaning
- When Does Contrastive Visual Representation Learning Work? (CVPR 2022 · similarity 0.89)
- MaskCLIP: Masked Self-Distillation Advances Contrastive Language-Image Pretraining (CVPR 2023 · similarity 0.88)
- A Simple Framework for Contrastive Learning of Visual Representations (ICML 2020 · similarity 0.88)
- Self-Organizing Visual Prototypes for Non-Parametric Representation Learning (ICML 2025 · similarity 0.88)
- Hybrid Distillation: Connecting Masked Autoencoders with Contrastive Learners (ICLR 2024 · similarity 0.87)
- A Theoretical Analysis of Self-Supervised Learning for Vision Transformers (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.
