Research map / Detection and segmentation / CNN features
Trapped in Texture Bias? A Large Scale Comparison of Deep Instance Segmentation
Johannes Theodoridis, Jessica Hofmann, Johannes Maucher, Andreas Schilling
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 |
|---|---|---|---|
| use deeper backbones (R101 vs R50, Swin-S vs Swin-T) (ablation) | instance segmentation frameworks with FPN neck · Stylized COCO | rAP | Improved |
| use Swin Transformer backbone instead of ResNet of similar complexity (ablation) | Mask R-CNN / Cascade · Stylized COCO | rAP | Improved |
| deeper/Swin backbones on small objects (ablation) | instance segmentation frameworks · Stylized COCO | rAPs | Mixed |
| use FPN neck instead of alternatives (C4, DC5, DCN) (ablation) | Mask R-CNN, Cascade, YOLO etc. · Stylized COCO | rAP | Hurt |
| add deformable convolutions (DCN) to the neck (ablation) | FPN models · Stylized COCO | rAP | Improved |
| use DC5 neck (ablation) | models on small objects · Stylized Objects | rAPs | Improved |
| change pre-training (supervised vs self-supervised), schedule or augmentation (ablation) | Mask R-CNN R50 · Stylized COCO | rAP | No clear effect |
| random initialization with Large Scale Jittering instead of ImageNet pre-training (ablation) | Mask R-CNN · Stylized COCO | rAP | Improved |
| restrict stylization to objects only instead of full image (ablation) | average framework · Stylized Objects vs Stylized COCO | rAP: 72 -> 47 vs 58 -> 18 | Improved |
| stylize background only (ablation) | average framework · Stylized Background | rAPs: 63 -> 63 | Hurt |
| stylize background only on large objects (ablation) | average framework · Stylized Background | rAPl: 96 -> 101 | No clear effect |
| use YOLACT++ framework instead of Mask R-CNN / Cascade / BMask / BCNet / DETR (ablation) | ResNet-50 + FPN frameworks · Stylized COCO | rAP | Improved |
Closest papers by meaning
- Training-Free Dataset Pruning for Instance Segmentation (ICLR 2025 · similarity 0.88)
- Pixelwise Instance Segmentation With a Dynamically Instantiated Network (CVPR 2017 · similarity 0.88)
- Fully Convolutional Instance-Aware Semantic Segmentation (CVPR 2017 · similarity 0.88)
- v-CLR: View-Consistent Learning for Open-World Instance Segmentation (CVPR 2025 · similarity 0.87)
- ImageNet-trained CNNs are not biased towards texture: Revisiting feature reliance through controlled suppression (NeurIPS 2025 · similarity 0.87)
- A Generative Approach at the Instance-Level for Image Segmentation Under Limited Training Data Conditions (Student Abstract) (AAAI 2025 · 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.
