Extreme R-CNN: Few-Shot Object Detection via Sample Synthesis and Knowledge Distillation
Shenyong Zhang, Wenmin Wang, Zhibing Wang, Honglei Li, Ruochen Li, Shixiong Zhang · 2024
Read Extreme R-CNN: Few-Shot Object Detection via Sample Synthesis and Knowledge Distillation by Shenyong Zhang, Wenmin Wang, Zhibing Wang, Honglei Li, Ruochen Li, Shixiong Zhang (2024) in the Tower of Babel, a free online library of public-domain works preserved in clean, searchable text. No sign-up. The full text is served as a plain-text file at a predictable URL, where text is present: https://antarctic-labs.com/tower-of-babel/pmc-11645051.txt.
Description
Shenyong Zhang and five co-authors present a 2024 paper on detecting objects in photos when only a few examples exist. Standard detectors need masses of labeled instances. Few-shot detectors trained on handfuls of new-class photos tend to favor the classes they already know and wobble on varied new samples. The team answers with Extreme R-CNN, a fine-tuning scheme that leans on made-up samples and teacher guidance.
They synthesize extra training images by clipping instances and augmenting them. The detector's design splits box drawing from class naming, so fake samples can train naming alone. A distillation step passes knowledge from a stronger model. On two standard benchmarks, scores for new classes rose by up to fifteen and six percent. In the one-shot test, their model beat a rival given ten shots. The work targets settings where labels are dear.
Catalog record
| CREATOR | Shenyong Zhang, Wenmin Wang, Zhibing Wang, Honglei Li, Ruochen Li, Shixiong Zhang |
|---|---|
| YEAR | 2024 |
| COLLECTION | PAPERS |
| FORMAT | TXT |
| FILE SIZE | 38 KB |
| VERSION | 1.0 |
| LICENSE | CC BY 4.0 |
| RIGHTS | CC BY |
| SOURCE | Europe PMC |