Research map / Detection and segmentation / CNN features
Scaling Object Detection by Transferring Classification Weights
Jason Kuen, Federico Perazzi, Zhe Lin, Jianming Zhang, Yap-Peng Tan
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 input normalization (ablation) | WTN · VG-200 | AR@100: 41.91 -> 54.19 | Improved |
| add group normalization (ablation) | WTN · VG-200 | AR@100: 41.91 -> 58.55 | Improved |
| add both input and group normalization (WTN+) (ablation) | WTN · OI-500 | AP50: 52.87 -> 58.82 | Improved |
| add both input and group normalization (WTN+) (ablation) | WTN · OI-57 | AP50: 34.94 -> 39.28 | Improved |
| add both input and group normalization (WTN+) (ablation) | WTN · VG-200 | AR@100: 41.91 -> 65.60 | Improved |
| add autoencoder reconstruction loss (AE-WTN) | WTN+ · OI-500 | AP50: 58.82 -> 59.59 | Improved |
| add autoencoder reconstruction loss (AE-WTN) | WTN+ · OI-57 | AP50: 39.28 -> 41.07 | Improved |
| add autoencoder reconstruction loss (AE-WTN) | WTN+ · VG-200 | AR@100: 65.60 -> 66.75 | Improved |
| add activity regularizer instead of reconstruction loss (ablation) | WTN+ · VG-200 | AR@100: 65.60 -> 36.26 | Hurt |
| add 5x weight decay (ablation) | WTN+ · OI-57 | AP50: 39.28 -> 40.79 | Improved |
| add Dropout (ablation) | WTN+ · OI-57 | AP50: 39.28 -> 40.09 | Improved |
| halve hidden channels (reduced capacity) (ablation) | WTN+ · OI-57 | AP50: 39.28 -> 37.81 | Hurt |
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 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
- Large Scale Semi-Supervised Object Detection Using Visual and Semantic Knowledge Transfer (CVPR 2016 · similarity 0.88)
- Revisiting Knowledge Transfer for Training Object Class Detectors (CVPR 2018 · similarity 0.88)
- Visual and Semantic Knowledge Transfer for Large Scale Semi-Supervised Object Detection (TPAMI 2017 · similarity 0.88)
- Scale-Transferrable Object Detection (CVPR 2018 · similarity 0.88)
- Boosting Weakly Supervised Object Detection with Progressive Knowledge Transfer (ECCV 2020 · similarity 0.88)
- CaT: Weakly Supervised Object Detection With Category Transfer (ICCV 2021 · 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.
