Weakly-Supervised Object Localization 벤치마크
Weakly-Supervised Object Localization on ImageNet
GT-known localization accuracy
- 2021-09-13 — CexCNN: GT-known localization accuracy 67.65
- 2022-07-21 — Deit-S: GT-known localization accuracy 68.8
- 2023-07-19 — Stable diffusion: GT-known localization accuracy 75.0
| Rank | Model | GT-known localization accuracy | Top-1 Localization Accuracy | average top-1 classification accuracy | Extra Training Data | Paper | Code | Year |
|---|---|---|---|---|---|---|---|---|
| 1 | Stable diffusion | 75.0 | 65.2 | – | ✓ | Generative Prompt Model for Weakly Supervised Object Localization | callsys/genpromp | 2023 |
| 2 | Deit-S | 68.8 | 56.1 | 76.7 | Weakly Supervised Object Localization via Transformer with Implicit Spatial Calibration | 164140757/scm · author31/scm_nbdev | 2022 | |
| 3 | CexCNN | 67.65 | – | – | Causal Explanation of Convolutional Neural Networks | HichemDebbi/CexCNN | 2021 | |
| 4 | TokenCut | 65.4 | 52.3 | – | Self-Supervised Transformers for Unsupervised Object Discovery using Normalized Cut | YangtaoWANG95/TokenCut | 2022 | |
| 5 | FALcon | 62.45 | 49.39 | – | Exploring Foveation and Saccade for Improved Weakly-Supervised Localization | TimurIbrayev/FALcon | 2023 | |
| 6 | R-Mix (ResNet-50) | – | 55.58 | – | Expeditious Saliency-guided Mix-up through Random Gradient Thresholding | minhlong94/random-mixup | 2022 |