Research map / Detection and segmentation / Weakly supervised detection and segmentation
SparseDet: Improving Sparsely Annotated Object Detection with Pseudo-positive Mining
Saksham Suri, Saketh Rambhatla, Rama Chellappa, Abhinav Shrivastava
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 pseudo-positive mining with self-supervised consistency loss instead of pseudo-labels alone | SparseDet · COCO Split-1 70% removed | AP: 24.93 -> 32.68 | Improved |
| add common RPN (C-RPN) over two views (ablation) | SparseDet · VOC07 Split-5 50% | AP: 42.67 -> 45.30 | Improved |
| add PPM on top of C-RPN without SSL loss (ablation) | SparseDet · VOC07 Split-5 50% | AP: 45.30 -> 45.44 | No clear effect |
| add SSL consistency loss together with PPM (ablation) | SparseDet · VOC07 Split-5 50% | AP: 45.44 -> 46.00 | Improved |
| add pseudo-labeling loss to SSL-based method (ablation) | SparseDet · VOC07 Split-5 50% | AP: 45.51 -> 46.00 | Improved |
| use single set of proposals from original image instead of C-RPN (ablation) | SparseDet · VOC07 Split-5 50% | AP: 46.00 -> 44.86 | Hurt |
| combine two separate proposal sets instead of C-RPN (ablation) | SparseDet · VOC07 Split-5 50% | AP: 46.00 -> 45.25 | Hurt |
| raise PPM objectness threshold to 0.95 (ablation) | SparseDet · VOC07 Split-5 50% | AP: 46.00 -> 45.30 | Hurt |
| use augmented view at inference (ablation) | SparseDet · COCO Splits 1-3 | AP: 38.32 -> 38.22 (Split-1 30%) | No clear effect |
| use FPN instead of C4 backbone (ablation) | SparseDet · COCO Splits 1-3 | AP: 37.67 -> 38.22 (Split-1 30%); 33.16 -> 32.68 (Split-1 70%) | Mixed |
| start PPM after warm-up of 9000 iterations instead of 0 (ablation) | SparseDet · COCO-mini ablation | AP: 45.51 -> 46.00 | Improved |
| add bounding box erase augmentation alone (ablation) | SparseDet · VOC07 Split-5 50% | AP: 42.76 -> 45.11 | Improved |
Later papers that compared with this method
4 later accepted papers report SparseDet as a baseline in their results tables; 4 of them report a better number on most of the results they share with it.
- MonoSAOD: Monocular 3D Object Detection with Sparsely Annotated Label (CVPR 2026 · better on 2 of 2 shared results)
- As Pseudo-Label Free as Possible: Leveraging Adaptive Feature Generation for Sparsely Annotated Object Detection (AAAI 2025 · better on 14 of 14 shared results)
- Learning Class Prototypes for Unified Sparse-Supervised 3D Object Detection (CVPR 2025 · better on 3 of 3 shared results)
- Multispectral Pedestrian Detection with Sparsely Annotated Label (AAAI 2025 · 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
- Co-Student: Collaborating Strong and Weak Students for Sparsely Annotated Object Detection (ECCV 2024 · similarity 0.91)
- Co-mining: Self-Supervised Learning for Sparsely Annotated Object Detection (AAAI 2021 · similarity 0.91)
- Calibrated Teacher for Sparsely Annotated Object Detection (AAAI 2023 · similarity 0.90)
- Label, Verify, Correct: A Simple Few Shot Object Detection Method (CVPR 2022 · similarity 0.89)
- As Pseudo-Label Free as Possible: Leveraging Adaptive Feature Generation for Sparsely Annotated Object Detection (AAAI 2025 · similarity 0.89)
- Object Detection Benchmarks are Incomplete: The Role of Label Errors and Annotation Uncertainty (NeurIPS 2026 · 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.
