Research map / Detection and segmentation / Open-set and few-shot detection
A Low-Shot Object Counting Network With Iterative Prototype Adaptation
Nikola Đukić, Alan Lukežič, Vitjan Zavrtanik, Matej Kristan
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 global self-attention block to the image encoder (ablation) | LOCA · FSC147 few-shot | MAE: 11.99 -> 10.24 (val MAE) | Improved |
| add the object prototype extraction (OPE) module instead of pooling exemplar features (ablation) | LOCA · FSC147 few-shot | MAE: 16.24 -> 10.24 (val MAE) | Improved |
| add exemplar shape queries to OPE (ablation) | LOCA · FSC147 few-shot | MAE: 13.77 -> 10.24 (val MAE) | Improved |
| compute shape queries from exemplar width and height instead of trainable queries (ablation) | LOCA · FSC147 few-shot | MAE: 13.00 -> 10.24 (val MAE) | Improved |
| use first cross-attention in OPE instead of simple summation of shape and appearance queries (ablation) | LOCA · FSC147 few-shot | MAE/RMSE: 5% MAE and 22% RMSE increase without it | Improved |
| set number of adaptation iterations L to 3 rather than 1 or 5 (ablation) | LOCA · FSC147 (val+test) | MAE: 11.07 (L=1), 11.60 (L=5) -> 10.50 (L=3) | Improved |
| use SwAV-pretrained backbone instead of ImageNet-pretrained (ablation) | LOCA · FSC147 few-shot | MAE: 8% MAE and 4% RMSE drop with ImageNet | Improved |
| increase input resolution from 384 to 512 (ablation) | LOCA · FSC147 few-shot | MAE/RMSE: 12.75 -> 10.79 (test MAE) | Improved |
| change prototype spatial size s from 3 to 1 or 5 (ablation) | LOCA · FSC147 few-shot | MAE/RMSE: test MAE 10.79 (s=1), 10.79 (s=3), 12.27 (s=5) | No clear effect |
| add object-count normalization to the l2 loss (ablation) | LOCA · FSC147 few-shot | MAE: 12.83 -> 10.79 (test MAE) | Improved |
| add auxiliary losses on intermediate OPE iterations (ablation) | LOCA · FSC147 few-shot | RMSE: 17% RMSE drop without it | Improved |
| extend LOCA to zero-shot with trainable objectness queries | LOCA · FSC147 zero-shot | MAE: RMSE 54.96 vs RCC 58.81; test MAE 16.22 vs CounTR 14.12 | Mixed |
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.
- Faster-RCNN — Faster R-CNN: Towards Real-Time Object Detection with Region Proposal Networks (TPAMI 2016) · 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 Novel Unified Architecture for Low-Shot Counting by Detection and Segmentation (NeurIPS 2024 · similarity 0.91)
- DAVE - A Detect-and-Verify Paradigm for Low-Shot Counting (CVPR 2024 · similarity 0.89)
- Single Domain Generalization for Few-Shot Counting via Universal Representation Matching (CVPR 2025 · similarity 0.89)
- Point Segment and Count: A Generalized Framework for Object Counting (CVPR 2024 · similarity 0.88)
- VLCounter: Text-Aware Visual Representation for Zero-Shot Object Counting (AAAI 2024 · similarity 0.88)
- T2ICount: Enhancing Cross-modal Understanding for Zero-Shot Counting (CVPR 2025 · similarity 0.88)
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.
