Xu Ma, Zhinan Qiao, Jingda Guo, Sihai Tang, Qi Chen, Qing Yang, Song Fu ✉️
(Accepted by ICIP 2020)
In this repository, all the models are implemented by pytorch.
We use the standard data augmentation strategies with ResNet.
To reproduce our CCD module work, please refer Usage.md.
😊 All trained models and training log files are submitted to Google Drive.
😊 We provide corresponding links in the "download" column.
Table 1: Comparison results of single-crop classification accuracy (%) and complexity on the ImageNet validation set. The best two performances are marked in **bold**.
| Model | top-1 acc. | top-5 acc. | FLOPs(G) | Parameters(M) | Download |
|---|---|---|---|---|---|
| ResNet50 | 75.8974 | 92.7224 | 4.122 | 25.557 | model log |
| SE-ResNet50 | 77.2877 | 93.6478 | 4.130 | 28.088 | model log |
| GE-ResNet50 | 76.2357 | 92.9847 | 4.127 | 25.557 | model log |
| CBAM-ResNet50 | 77.2840 | 93.6005 | 4.139 | 28.090 | model log |
| SK-ResNet50 | 77.3657 | 93.5256 | 4.187 | 26.154 | model log |
| GC-ResNet50 | 74.8966 | 92.2812 | 4.130 | 28.105 | model log |
| CCD-ResNet50 (ours) | 77.3137 | 93.6489 | 4.122 | 25.560 | model log |
Table 2: Detection performances (%) with different backbones on the MS-COCO validation dataset. We employ two state-of-the-art detectors: RetinaNet and Cascade R-CNN in our detection experiments.
| Detector | Backbone | AP(50:95) | AP(50) | AP(75) | AP(s) | AP(m) | AP(l) | Download |
|---|---|---|---|---|---|---|---|---|
| Retina | ResNet50 | 36.2 | 55.9 | 38.5 | 19.4 | 39.8 | 48.3 | model log |
| Retina | SE-ResNet50 | 37.4 | 57.8 | 39.8 | 20.6 | 40.8 | 50.3 | model log |
| Retina | CCD-ResNet50 | 37.8 | 58.5 | 40.1 | 21.6 | 41.5 | 50.9 | model log |
| Cascade R-CNN | ResNet50 | 40.6 | 58.9 | 44.2 | 22.4 | 43.7 | 54.7 | model log |
| Cascade R-CNN | GC-ResNet50 | 41.1 | 59.7 | 44.6 | 23.6 | 44.1 | 54.3 | model log |
| Cascade R-CNN | CCD-ResNet50 | 42.5 | 61.1 | 46.4 | 24.7 | 45.9 | 56.5 | model log |