ParitoshParmar / HalluciNet--PyTorch

Approximating a 3DCNN with a 2DCNN

HalluciNet

HalluciNet-ing Spatiotemporal Representations Using 2D-CNN

HalluciNet Concept:

Here, we have used VGG-bn-11 and ResNeXt3d-101 as our 2D-CNN student and 3D-CNN teacher, respectively. You may experiment with any other student-teacher combination.

If you find this code useful, please consider citing:

@article{hallucinet,
  title={HalluciNet-ing Spatiotemporal Representations Using 2D-CNN},
  author={Parmar, Paritosh and Morris, Brendan},
  journal={arXiv preprint arXiv:1912.04430},
  year={2019}
}