search

Wizaron / instance-segmentation-pytorch

Semantic Instance Segmentation with a Discriminative Loss Function in PyTorch
https://arxiv.org/abs/1708.02551
GNU General Public License v3.0
458 stars 95 forks source link

There are some problems in the implementation of 'recurrent_hourglass.py'. #13

Open sunyasheng opened 5 years ago

sunyasheng commented 5 years ago

In the implementation of 'recurrent_hourglass.py'. There is only oneConvGRUCell

self.convgru_cell = ConvGRUCell

But it seems that there are supposed to be n_levels ConvGRUCell, each of which tackles one type of resolution feature maps. As for the recurrent part, I think you can just maintain the hidden state of GRUCell and let the input at next timestep come in.

That's my understanding of the paper Instance Segmentation and Tracking with Cosine Embeddings and Recurrent Hourglass Networks. If you think in another way, let me know.