Research map / Detection and segmentation / Open-set and few-shot detection
Bamboo: Building Mega-Scale Vision Dataset Continually with Human–Machine Synergy
Yuanhan Zhang, Qinghong Sun, Yichun Zhou, Zexin He, Zhenfei Yin, Kun Wang, Lu Sheng, Yu Qiao, Jing Shao, Ziwei 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 |
|---|---|---|---|
| pretrain on Bamboo-CLS instead of CLIP WIT 400M | ViT-B/16 · downstream linear probe average | accuracy: 85.6 -> 91.8 | Improved |
| pretrain on Bamboo-CLS instead of CLIP WIT 400M | RN50 · downstream linear probe average | accuracy: 79.1 -> 86.0 | Improved |
| pretrain on Bamboo-DET instead of Objects365 | RN50-FPN · PASCAL VOC | AP50: 86.4 -> 87.5 | Improved |
| pretrain on Bamboo-DET instead of Objects365 | RN50-FPN · CityPersons | MR: 14.7 -> 12.6 | Improved |
| pretrain on Bamboo-DET instead of Objects365 | RN50-FPN · COCO | mmAP: 39.3 -> 43.9 | Improved |
| use Bamboo-CLS pretraining instead of CLIP | RN50 · zero-shot classification average | accuracy: 70.0 -> 76.0 | Improved |
| fine-tune Bamboo-CLS instead of IG-1B SWSL | RN50 · fine-tuning average | accuracy: 86.8 -> 87.3 | Improved |
| fine-tune Bamboo-CLS instead of ImageNet22K BiT-M | RN50 · fine-tuning average | accuracy: 86.0 -> 87.3 | Improved |
| pretrain on Bamboo instead of JFT-300M | RN50 · ObjectNet | accuracy: 37.6 -> 38.8 | Improved |
| pretrain on Bamboo instead of ANN-1.3B | ViT-B/16 · ObjectNet | accuracy: 50.7 -> 53.9 | Improved |
| label sparse images instead of crowded images under same proposal budget (ablation) | Objects365 subsets · PASCAL VOC | AP50: 85.1 -> 86.3 | Improved |
| sample randomly instead of keeping dense crowded images (same budget) (ablation) | Objects365 subsets · PASCAL VOC | AP50: 85.1 -> 85.8 | Improved |
Closest papers by meaning
- SELECT: A Large-Scale Benchmark of Data Curation Strategies for Image Classification (NeurIPS 2024 · similarity 0.86)
- Similarity Search for Efficient Active Learning and Search of Rare Concepts (AAAI 2022 · similarity 0.86)
- Simple Multi-Dataset Detection (CVPR 2022 · similarity 0.85)
- Active Learning Meets Foundation Models: Fast Remote Sensing Data Annotation for Object Detection (ICCV 2025 · similarity 0.85)
- COCONut: Modernizing COCO Segmentation (CVPR 2024 · similarity 0.84)
- A Cross-Domain Benchmark for Active Learning (NeurIPS 2024 · similarity 0.84)
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.
