Research map / Detection and segmentation / Weakly supervised detection and segmentation
Data Distillation: Towards Omni-Supervised Learning
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 multi-transform distilled labels from un-120 | ResNet-50 keypoint detector trained on co-115 · COCO val2017 keypoint detection | AP: 65.1 -> 67.1 | Helped |
| Add multi-transform distilled labels from s1m-180 | ResNeXt-101-64×4 keypoint detector trained on co-115 · COCO val2017 keypoint detection | AP: 67.3 -> 68.5 | Helped |
| Add multi-transform distilled labels from un-120 | ResNeXt-101-32×4 object detector trained on co-115 · COCO val2017 object detection | AP: 40.1 -> 41.0 | Helped |
| Use co-80 as unlabeled data with co-35 labeled data | ResNet-50 keypoint detector · COCO val2017 keypoint detection | AP: 54.9 -> 60.2 | Helped |
| Use co-80 as unlabeled data with co-35 labeled data | ResNet-50 object detector · COCO val2017 object detection | AP: 30.5 -> 32.3 | Helped |
| Add test-time augmentations (ablation) | Data-distilled keypoint detectors · COCO val2017 keypoint detection | AP: 67.1 -> 68.9 | Helped |
| Increase fully supervised training from 130k to 270k iterations (ablation) | ResNet-50 keypoint detector trained on co-115 · COCO keypoint detection | AP: 65.1 -> 64.7 | Hurt |
| Increase data-distillation training from 90k to 360k iterations (ablation) | ResNet-50 keypoint detector trained on co-115 plus s1m-180 · COCO keypoint detection | AP: 63.6 -> 66.6 | Helped |
| Increase the amount of unlabeled data (ablation) | Data-distilled keypoint detector · COCO keypoint detection, co-115 plus a fraction of un-120 | AP | Helped |
| Use a higher-quality teacher (ablation) | Data-distilled ResNet-50 keypoint detector · COCO keypoint detection, co-115 plus un-120 | AP | Helped |
| Use multi-transform inference for pseudo-label generation | Data-distillation keypoint detector · COCO keypoint detection | AP: 65.1 -> 67.8 | Helped |
Closest papers by meaning
- Omni-DETR: Omni-Supervised Object Detection With Transformers (CVPR 2022 · similarity 0.87)
- Scale-Equivalent Distillation for Semi-Supervised Object Detection (CVPR 2022 · similarity 0.86)
- Boost Self-Supervised Dataset Distillation via Parameterization, Predefined Augmentation, and Approximation (ICLR 2025 · similarity 0.86)
- Dataset Distillation for Pre-Trained Self-Supervised Vision Models (NeurIPS 2025 · similarity 0.86)
- DetKDS: Knowledge Distillation Search for Object Detectors (ICML 2024 · similarity 0.85)
- Towards Universal Dataset Distillation via Task-Driven Diffusion (CVPR 2025 · similarity 0.84)
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.