This is an unofficial implementation of ECCV 18 paper "SphereNet: Learning Spherical Representations for Detection and Classification in Omnidirectional Images".

Currently only 3x3 SphereNet's Conv2D and MaxPool2D are implemented. For now only mode=bilinear is allowed, mode=nearest have to wait until Pytorch1.0. You can replace any model's CNN with SphereNet's CNN, they are implemented such that you can directly load pretrained weight to SphereNet's CNN.

• python3
• pytorch>=0.4.1
• numpy
• scipy

Copy spherenet directory to your project.
If you want to install as an package such that you can import spherenet everywhere:

cd $YOUR_CLONED_SPHERENET_DIR
pip install .

from spherenet import SphereConv2D, SphereMaxPool2D

conv1 = SphereConv2D(1, 32, stride=1)
pool1 = SphereMaxPool2D(stride=2)

# toy example
img = torch.randn(1, 1, 60, 60)  # (batch, channel, height, weight)
out = conv1(img)  # (1, 32, 60, 60)
out = pool1(out)  # (1, 32, 30, 30)

• To apply SphereNet in your trained model, simply replace the nn.Conv2d with SphereConv2D, and replace nn.MaxPool2d with SphereMaxPool2D. They should work well with load_state_dict.

• Classification OminiMNIST data (spherenet.OmniMNIST, spherenet.OmniFashionMNIST)
  • 
• Reproduce OmniMNIST Result

  • Method	Test Error (%)
    SphereNet ( paper )	5.59
    SphereNet ( ours )	5.77
    EquirectCNN ( paper )	9.61
    EquirectCNN ( ours )	9.63

  • 

• paper
  • Benjamin Coors, Alexandru Paul Condurache, Andreas Geiger
  • ECCV2018

      @inproceedings{coors2018spherenet,
        title={SphereNet: Learning Spherical Representations for Detection and Classification in Omnidirectional Images},
        author={Coors, Benjamin and Condurache, Alexandru Paul and Geiger, Andreas},
        booktitle={Proceedings of the European Conference on Computer Vision (ECCV)},
        pages={518--533},
        year={2018}
      }