Research map / Architectures and efficiency / RNNs and normalization
Partial Weight Adaptation for Robust DNN Inference
Xiufeng Xie, Kyu-Han Kim
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 |
|---|---|---|---|
| Adapt distortion-sensitive weights for each JPEG quality instead of using one fixed model | DRN-D-22 · Cityscapes, 10 JPEG qualities | average accuracy: 18.12% | Improved |
| Adapt distortion-sensitive weights for each JPEG quality instead of mixed training | DRN-D-22 · Cityscapes, 10 JPEG qualities | average accuracy: 1.95% | Improved |
| Adapt distortion-sensitive weights for each resolution instead of using one fixed model | DRN-D-38 · Cityscapes, 7 H.264 resolutions | average accuracy: 2.88% | Improved |
| Adapt distortion-sensitive weights for each resolution instead of mixed training | DRN-D-38 · Cityscapes, 7 H.264 resolutions | average accuracy: 1.70% | Improved |
| Adapt distortion-sensitive weights for each brightness level instead of using one fixed model | DRN-D-38 · Cityscapes, 20 brightness levels | average accuracy: 5.11% | Improved |
| Adapt distortion-sensitive weights for each brightness level instead of mixed training | DRN-D-38 · Cityscapes, 20 brightness levels | average accuracy: 1.11% | Improved |
| Adapt distortion-sensitive weights for each JPEG quality instead of using one fixed model | Mask R-CNN with ResNet-50 · Penn-Fudan, 10 JPEG qualities, pedestrian segmentation | average accuracy: 5.16% | Improved |
| Adapt distortion-sensitive weights for each JPEG quality instead of using one fixed model | Mask R-CNN with ResNet-50 · Penn-Fudan, 10 JPEG qualities, bounding box detection | average accuracy: 3.53% | Improved |
| Adapt distortion-sensitive weights for each JPEG quality instead of mixed training | Mask R-CNN with ResNet-50 · Penn-Fudan, 10 JPEG qualities, pedestrian segmentation | average accuracy: 1.20% | Improved |
| Adapt distortion-sensitive weights for each JPEG quality instead of mixed training | Mask R-CNN with ResNet-50 · Penn-Fudan, 10 JPEG qualities, bounding box detection | average accuracy: 1.33% | Improved |
| Adapt distortion-sensitive weights for each H.264 distortion level rather than use stability training | DRN-D-38 · Cityscapes, H.264 compression | average accuracy: 4.84% | Improved |
| Adapt only 0.2% of DRN weights rather than fully fine-tune the model (ablation) | DRN-D-38 · Distorted-data frequency response | frequency response similarity: 0.2% | No clear effect |
Closest papers by meaning
- Bandwidth-Aware Adaptive Codec for DNN Inference Offloading in IoT (ECCV 2022 · similarity 0.86)
- Instant Video Models: Universal Adapters for Stabilizing Image-Based Networks (NeurIPS 2025 · similarity 0.85)
- Dynamic Tuning Towards Parameter and Inference Efficiency for ViT Adaptation (NeurIPS 2024 · similarity 0.84)
- E4: Energy-Efficient DNN Inference for Edge Video Analytics via Early Exiting and DVFS (AAAI 2025 · similarity 0.83)
- Make Me a BNN: A Simple Strategy for Estimating Bayesian Uncertainty from Pre-trained Models (CVPR 2024 · similarity 0.82)
- Verification of Neural Networks Against Convolutional Perturbations via Parameterised Kernels (AAAI 2025 · similarity 0.82)
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.
