Research map / Vision-language and multimodal / Cross-modal learning
Cross Modal Distillation for Supervision Transfer
Check 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 |
|---|---|---|---|
| train depth features by supervision transfer rather than copying RGB weights | AlexNet depth detector · NYU val | mean AP: 25.1 -> 29.7 | Helped |
| use VGG rather than AlexNet as the supervising network (ablation) | Depth representation · NYU val | mean AP: 30.0 -> 32.2; 30.5 -> 33.6 | Helped |
| add depth features learned by supervision transfer to RGB features | NYU detection · NYU val | mean AP: 22.3 -> 35.6 | Helped |
| add pool2 and conv4 hyper-column features to fc7 (ablation) | Supervision-transfer depth network · NYU val | regionAP at 0.5: 28.4 -> 31.5 | Helped |
| transfer supervision at pool1 or pool2 rather than a mid-level layer (ablation) | Depth detector · NYU val | boxAP: 24.4; 28.4 versus 30.5 | Hurt |
| transfer supervision at fc6 or fc7 rather than a mid-level layer (ablation) | Depth detector · NYU val | boxAP: 29.7; 27.7 versus 30.5 | Hurt |
| transfer at conv3 and fc7 together instead of either layer alone (ablation) | Depth detector · NYU val | boxAP | Helped |
| use disparity images instead of HHA input (ablation) | Supervision-transfer depth detector · NYU val | mAP: 30.5 -> 29.2 | Hurt |
| use supervision transfer instead of random initialization | Optical-flow action detector · JHMDB test | mean AP: 31.7 -> 35.7 | Helped |
| use supervision transfer instead of the prior depth feature initialization | Depth-only detector · NYU test | mean AP: 34.2 -> 41.7 | Helped |
| use supervision transfer for depth features instead of prior features | RGB + depth detector with AlexNet RGB · NYU test | mean AP: 44.4 -> 47.1 | Helped |
| use supervision transfer for depth features instead of prior features | RGB + depth detector with VGG RGB · NYU test | mean AP: 46.2 -> 49.1 | Helped |
Closest papers by meaning
- Local-Global Multi-Modal Distillation for Weakly-Supervised Temporal Video Grounding (AAAI 2024 · similarity 0.86)
- C2KD: Bridging the Modality Gap for Cross-Modal Knowledge Distillation (CVPR 2024 · similarity 0.85)
- Connect, Collapse, Corrupt: Learning Cross-Modal Tasks with Uni-Modal Data (ICLR 2024 · similarity 0.85)
- XKD: Cross-Modal Knowledge Distillation with Domain Alignment for Video Representation Learning (AAAI 2024 · similarity 0.85)
- Beyond Modality Collapse: Representation Blending for Multimodal Dataset Distillation (NeurIPS 2025 · similarity 0.84)
- Flowing from Words to Pixels: A Noise-Free Framework for Cross-Modality Evolution (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.