Skip to content

Latest commit

 

History

History
51 lines (43 loc) · 2.88 KB

rednet.md

File metadata and controls

51 lines (43 loc) · 2.88 KB
Raw

RedNet

  • Paper：Involution: Inverting the Inherence of Convolution for Visual Recognition

  • Origin Repo：d-li14/involution

  • Code：rednet.py

  • Evaluate Transforms：

    # backend: cv2
# input_size: 224x224
transforms = T.Compose([
    T.Resize(256),
    T.CenterCrop(224),
    T.Normalize(
        mean=[123.675, 116.28, 103.53],
        std=[58.395, 57.12, 57.375],
        to_rgb=True,
        data_format='HWC'
    ),
    T.ToTensor(),
])

  • Model Details：

    Model Model Name Params (M) FLOPs (G) Top-1 (%) Top-5 (%) Pretrained Model
    RedNet-26 rednet_26 9.2 1.7 75.96 93.19 Download
    RedNet-38 rednet_38 12.4 2.2 77.48 93.57 Download
    RedNet-50 rednet_50 15.5 2.7 78.35 94.18 Download
    RedNet-101 rednet_101 25.7 4.7 78.92 94.35 Download
    RedNet-152 rednet_152 34.0 6.8 79.12 94.38 Download

  • Citation：

    @InProceedings{Li_2021_CVPR,
    title = {Involution: Inverting the Inherence of Convolution for Visual Recognition},
    author = {Li, Duo and Hu, Jie and Wang, Changhu and Li, Xiangtai and She, Qi and Zhu, Lei and Zhang, Tong and Chen, Qifeng},
    booktitle = {IEEE/CVF Conference on Computer Vision and Pattern Recognition (CVPR)},
    month = {June},
    year = {2021}
}