Research map / Detection and segmentation / CNN features
When Pixel Difference Patterns Meet ViT: PiDiViT for Few-Shot Object Detection
Hongliang Zhou, Yongxiang Liu, Canyu Mo, Weijie Li, Bowen Peng, Li Liu
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 |
|---|---|---|---|
| add MFFM (ablation) | DE-ViT · COCO FSOD ablation (T=1) | nAP50: 49.4 -> 51.6 | Improved |
| add DCFM (ablation) | DE-ViT · COCO FSOD ablation (T=1) | nAP50: 49.4 -> 51.4 | Improved |
| add cosine-similarity projection subspace CSPS (ablation) | DE-ViT · COCO FSOD ablation (T=1) | nAP50: 49.4 -> 50.1 | Improved |
| combine MFFM, DCFM and CSPS (ablation) | DE-ViT · COCO FSOD ablation (T=1) | nAP50: 49.4 -> 54.1 | Improved |
| fuse APDC with standard convolution in DCFM (ablation) | DE-ViT · COCO FSOD ablation (T=1) | nAP50: 49.4 -> 49.0 | Hurt |
| fuse RPDC with standard convolution in DCFM (ablation) | DE-ViT · COCO FSOD ablation (T=1) | nAP50: 49.4 -> 49.4 | No clear effect |
| fuse CPDC with standard convolution in DCFM (ablation) | DE-ViT · COCO FSOD ablation (T=1) | nAP50: 49.4 -> 51.4 | Improved |
| use attention fusion instead of static fusion in MFFM (ablation) | MFFM · COCO FSOD ablation (T=1) | nAP50: 50.7 -> 51.6 | Improved |
| use dilation 2 instead of 4 for dilated 3x3 branch (ablation) | MFFM · COCO FSOD ablation (T=1) | nAP50: 51.2 -> 51.6 | Improved |
| add 7x7 and dilated 3x3 branches to MFFM kernels (ablation) | MFFM · COCO FSOD ablation (T=1) | nAP50: 50.7 -> 51.6 | Improved |
| replace DE-ViT dot-product with PiDiViT modules vs CD-ViTO | PiDiViT · COCO FSOD 10-shot (novel) | nAP50: 54.9 -> 57.6 | Improved |
| replace DE-ViT with PiDiViT in one-shot detection | PiDiViT · COCO one-shot (4 splits average) | nAP50: 28.4 -> 32.8 | Improved |
Earlier methods this paper compares with
Named baselines from earlier accepted papers in this paper's results tables: better than 1 of 1 on most shared results.
- CORA+ — CORA: Adapting CLIP for Open-Vocabulary Detection With Region Prompting and Anchor Pre-Matching (CVPR 2023) · better on 1 of 1 shared result
Matched by method name in results tables read by AI. Authors' own tables tend to favour the newer method, and a comparison can be missed when a method is renamed.
Closest papers by meaning
- A Simple Single-Scale Vision Transformer for Object Detection and Instance Segmentation (ECCV 2022 · similarity 0.87)
- Exploring Plain Vision Transformer Backbones for Object Detection (ECCV 2022 · similarity 0.87)
- Temporal Object-Aware Vision Transformer for Few-Shot Video Object Detection (AAAI 2026 · similarity 0.86)
- Context-Transformer: Tackling Object Confusion for Few-Shot Detection (AAAI 2020 · similarity 0.86)
- Self-Promoted Supervision for Few-Shot Transformer (ECCV 2022 · similarity 0.86)
- Strong Baselines for Parameter-Efficient Few-Shot Fine-Tuning (AAAI 2024 · similarity 0.85)
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.
