nervanasystems/caffe2neon

Tools to convert Caffe models to neon's serialization format

caffe2neon's Issues

How can I convert SegNet to neon format?

I'm trying to convert SegNet to neon, however it seems that some layer types like DenseImageData or Upsample are not supported. Is this a neon limitation?

• If yes, then are these layer types going to be supported by neon? When?

• If no, are you going support these types in caffe2neon soon or how can I help to extend the caffe2neon to support SegNet conversion?

Resnet 50

Hi,

When testing the tests/resnet/test_resnet.py with ResNet-50 I get:
Traceback (most recent call last): File "tests/resnet/test_resnet.py", line 119, in <module> model.initialize(train) File "/home/standard/neon/neon/models/model.py", line 128, in initialize self.layers.allocate() File "/home/standard/neon/neon/layers/container.py", line 295, in allocate l.allocate() File "/home/standard/neon/neon/layers/container.py", line 634, in allocate l.allocate(shared_outputs=self.outputs) File "/home/standard/neon/neon/layers/container.py", line 293, in allocate alloc_layers[-1].allocate(shared_outputs) File "/home/standard/neon/neon/layers/layer.py", line 129, in allocate parallelism=self.parallelism) File "/home/standard/neon/neon/backends/backend.py", line 587, in iobuf out_tsr = shared if shared.shape == bufshape else shared.share(bufshape) File "/home/standard/neon/neon/backends/nervanagpu.py", line 595, in share raise ValueError("total size of new array must <= size of parent") ValueError: total size of new array must <= size of parent

Am I doing something wrong?
Thanks