Research map / Detection and segmentation / Open-set and few-shot detection
On the Importance of Visual Context for Data Augmentation in Scene Understanding
Nikita Dvornik, Julien Mairal, Cordelia Schmid
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 |
|---|---|---|---|
| paste objects at random locations instead of no copy-paste (ablation) | Base-DA · VOC07-test single-category | average AP (%): 58.0 -> 56.9 | Hurt |
| add context-driven placement over random placement (ablation) | copy-paste DA · VOC07-test single-category | average AP (%): 56.9 -> 64.0 | Improved |
| add context-driven placement to standard augmentation (ablation) | Base-DA · VOC07-test single-category | average AP (%): 58.0 -> 64.0 | Improved |
| add context-driven augmentation, multi-category (ablation) | BlitzNet300 · VOC07-test multi-category | mAP (%): 64.6 -> 66.5 | Improved |
| add context-driven augmentation, multi-category (ablation) | Faster R-CNN · VOC07-test multi-category | mAP (%): 63.1 -> 64.5 | Improved |
| add context-driven augmentation for segmentation (ablation) | BlitzNet300 · VOC12val-seg | mean IoU (%): 63.3 -> 65.4 | Improved |
| random placement for segmentation (ablation) | BlitzNet300 · VOC12val-seg | mean IoU (%): 63.3 -> 63.1 | Hurt |
| remove object context (objects only in random context) (ablation) | single-category detector · VOC12 first 5 categories | average AP (%): 48.7 -> 33.9 | Hurt |
| enlarge and reblend objects in same location (right context) (ablation) | single-category detector · VOC12 first 5 categories | average AP (%): 48.7 -> 51.6 | Improved |
| add context-driven augmentation on COCO | BlitzNet300 · COCO val2017 detection | mAP@0.5:0.95 (%): 27.3 -> 28.0 | Improved |
| random placement on COCO (ablation) | Mask R-CNN · COCO val2017 detection | mAP@0.5:0.95 (%): 38.6 -> 36.9 | Hurt |
| add context-driven augmentation with decaying probability | Mask R-CNN · COCO val2017 detection | mAP@0.5:0.95 (%): 38.6 -> 39.1 | Improved |
Closest papers by meaning
- Modeling Visual Context is Key to Augmenting Object Detection Datasets (ECCV 2018 · similarity 0.95)
- UniT: Unified Knowledge Transfer for Any-Shot Object Detection and Segmentation (CVPR 2021 · similarity 0.88)
- The Devil is in the Object Boundary: Towards Annotation-free Instance Segmentation using Foundation Models (ICLR 2024 · similarity 0.88)
- Object-level Scene Context Prediction (TPAMI 2021 · similarity 0.88)
- Learning to Segment Every Thing (CVPR 2018 · similarity 0.88)
- Semantic-aware SAM for Point-Prompted Instance Segmentation (CVPR 2024 · 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.
