Research map / Detection and segmentation / Open-set and few-shot detection
MosaicOS: A Simple and Effective Use of Object-Centric Images for Long-Tailed Object Detection
Cheng Zhang, Tai-Yu Pan, Yandong Li, Hexiang Hu, Dong Xuan, Soravit Changpinyo, Boqing Gong, Wei-Lun Chao
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 2x2 mosaicking of object-centric images (ablation) | MosaicOS (two-stage, F) · LVIS v0.5 val | AP^b_r: 16.97 -> 18.76 | Improved |
| stitch images from different classes instead of the same class (ablation) | MosaicOS · LVIS v0.5 val | AP^b_r: 18.76 -> 19.73 | Improved |
| use two-stage fine-tuning instead of single-stage joint fine-tuning (ablation) | MosaicOS · LVIS v0.5 val | AP^b: 20.09 -> 24.75 | Improved |
| replace detector-predicted labels with the image class label on pseudo boxes (ablation) | MosaicOS (pre-trained detector pseudo-labels) · LVIS v0.5 val | AP^b_r: 13.93 -> 17.45 | Improved |
| add image class labels to calibrated-detector pseudo-boxes (ablation) | MosaicOS (calibrated detector) · LVIS v0.5 val | AP^b_r: 13.13 -> 19.31 | Improved |
| add 100 Google images per class to ImageNet object-centric images | MosaicOS · LVIS v0.5 val | AP^b_r: 19.73 -> 20.25 | Improved |
| optimize RPN and box regression losses in addition to classification loss in stage one (ablation) | MosaicOS · LVIS v0.5 val | AP^b_r: 18.87 -> 19.73 | Improved |
| use LORE pseudo-boxes instead of six fixed locations (two-stage) (ablation) | MosaicOS · LVIS v0.5 val | AP^b_r: 19.73 -> 20.06 | No clear effect |
| use 3x3 instead of 2x2 mosaic (ablation) | MosaicOS (same-class mosaic) · LVIS v0.5 val | AP^b_r: 18.76 -> 18.14 | No clear effect |
| stitch object-centric images instead of stitching scene-centric LVIS images (ablation) | Faster R-CNN* · LVIS v0.5 val | AP^b_r: 13.99 -> 19.73 | Improved |
| use ImageNet object-centric images instead of crawled Flickr images | MosaicOS · LVIS v0.5 val | AP^b_r: 16.17 -> 19.73 | Improved |
| add MosaicOS fine-tuning on classes already seen in ImageNet-1K backbone pre-training | Faster R-CNN* · LVIS v0.5 val (176 overlapping classes) | AP^b_r: 14.78 -> 21.16 | Improved |
Later papers that compared with this method
2 later accepted papers report MosaicOS as a baseline in their results tables; 2 of them report a better number on most of the results they share with it.
- Learning with Free Object Segments for Long-Tailed Instance Segmentation (ECCV 2022 · better on 4 of 4 shared results)
- On Model Calibration for Long-Tailed Object Detection and Instance Segmentation (NeurIPS 2021 · better on 3 of 3 shared results)
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 R-CNN — Faster R-CNN: Towards Real-Time Object Detection with Region Proposal Networks (TPAMI 2016) · better on 6 of 6 shared results
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
- SimLTD: Simple Supervised and Semi-Supervised Long-Tailed Object Detection (CVPR 2025 · similarity 0.90)
- Learning with Free Object Segments for Long-Tailed Instance Segmentation (ECCV 2022 · similarity 0.89)
- "A Simple Approach and Benchmark for 21,000-Category Object Detection" (ECCV 2022 · similarity 0.88)
- Towards Large-Scale Small Object Detection: Survey and Benchmarks (TPAMI 2023 · similarity 0.88)
- ObjectNet: A large-scale bias-controlled dataset for pushing the limits of object recognition models (NeurIPS 2019 · similarity 0.88)
- Deep Learning for Generic Object Detection: A Survey (IJCV 2019 · 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.
