Research map / Detection and segmentation / CNN features
Learning With Side Information Through Modality Hallucination
Judy Hoffman, Saurabh Gupta, Trevor Darrell
Check 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 a depth-supervised hallucination branch to the RGB detector | AlexNet Fast R-CNN · NYUD2 test | mAP: 26.6 -> 30.5 | Improved |
| Add a depth-supervised VGG-1024 hallucination branch | AlexNet RGB detector with VGG-1024 hallucination network · NYUD2 test | mAP: 29.9 -> 34.0 | Improved |
| Use a depth-initialized rather than randomly initialized hallucination network (ablation) | AlexNet hallucination detector · NYUD2 val | mAP: 20.3 -> 24.2 | Improved |
| Use depth-network rather than RGB-network initialization (ablation) | AlexNet hallucination detector · NYUD2 val | mAP: 22.4 -> 24.2 | Improved |
| Place the hallucination loss at pool5 instead of pool1 (ablation) | AlexNet hallucination detector · NYUD2 val | mAP: 22.3 -> 24.2 | Improved |
| Place the hallucination loss at pool5 instead of pool2 (ablation) | AlexNet hallucination detector · NYUD2 val | mAP: 21.5 -> 24.2 | Improved |
| Place the hallucination loss at pool5 instead of fc6 (ablation) | AlexNet hallucination detector · NYUD2 val | mAP: 20.4 -> 24.2 | Improved |
| Add the NYUD2-trained hallucination branch | RGB detector applied to PASCAL VOC 2007 · PASCAL VOC 2007 test | mAP: 16.9 -> 20.8 | Improved |
| Add the NYUD2-trained hallucination branch after RGB fine-tuning on PASCAL | PASCAL-fine-tuned RGB detector · PASCAL VOC 2007 test | mAP: 52.1 -> 53.4 | Improved |
| Use depth-estimated pixels as input to the depth channel instead of feature hallucination | Pre-trained RGB-D detector · NYUD2 | mAP: 22% mAP vs 27% mAP | Hurt |
| Combine two RGB models rather than using a single RGB model | RGB detector · NYUD2 test | mAP: 26.6 -> 29.0 | Improved |
Closest papers by meaning
- Analyzing and Mitigating Object Hallucination: A Training Bias Perspective (AAAI 2026 · similarity 0.88)
- Beyond the Global Scores: Fine-Grained Token Grounding as a Robust Detector of LVLM Hallucinations (CVPR 2026 · similarity 0.87)
- GLSim: Detecting Object Hallucinations in LVLMs via Global-Local Similarity (NeurIPS 2025 · similarity 0.87)
- Detecting and Preventing Hallucinations in Large Vision Language Models (AAAI 2024 · similarity 0.86)
- GHOST: Hallucination-Inducing Image Generation for Multimodal LLMs (ICLR 2026 · similarity 0.86)
- Same Attention, Different Truths: Put Logit-Lens over Visual Attention to Detect and Mitigate LVLM Object Hallucination (CVPR 2026 · similarity 0.86)
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.
