[WIP] Add SeparableConv1D Layer #157
Conversation
� Conflicts: � api/src/main/kotlin/org/jetbrains/kotlinx/dl/api/inference/keras/ModelSaver.kt
- still have to write tests, and do the 1D variant - this also fixed a bug in WeightLoader that did not load the bias weights correctly for SeparableConv2D.kt
|
This is still a work in progress but I would appreciate some feedback already if anyone has time :). Will write some tests and implement the 1D version soon. |
dosier
changed the title
| * | ||
| * Note: layer attributes cannot be modified after the layer has been called once (except the `trainable` attribute). | ||
| * | ||
| * TODO: add rank for getting the channel axis? |
There was a problem hiding this comment.
We are supporting at this moment only Channels Last (NWHC)
| * Note: layer attributes cannot be modified after the layer has been called once (except the `trainable` attribute). | ||
| * | ||
| * TODO: add rank for getting the channel axis? | ||
| * TODO: add docs for params? |
There was a problem hiding this comment.
Of course, we need docs for params
| * | ||
| * TODO: add rank for getting the channel axis? | ||
| * TODO: add docs for params? | ||
| * TODO: add trainable param? |
There was a problem hiding this comment.
What do you mean here? Please explain or provide an example?
| protected open fun getOutputDepth(numberOfChannels: Long): Long = filtersInternal | ||
|
|
||
| private fun computeMatricesShapes(numberOfChannels: Long) { | ||
| depthwiseKernelShape = shapeFromDims(*kernelSizeInternal, numberOfChannels, depthMulitplierInternal.toLong()) |
There was a problem hiding this comment.
maybe better to make the function without side-effects, which returns a triple of shapes
| @@ -32,13 +32,13 @@ internal fun depthwiseConv2dBiasVarName(name: String) = name + "_" + "depthwise_ | |||
| internal fun depthwiseConv2dKernelVarName(name: String) = name + "_" + "depthwise_conv2d_kernel" | |||
|
|
|||
| /** Default SeparableConv2d bias variable name in TensorFlow graph, based on variable's name. */ | |||
| internal fun separableConv2dBiasVarName(name: String) = name + "_" + "separable_conv2d_bias" | |||
| internal fun separableConvBiasVarName(name: String, dim: Int) = name + "_" + "separable_conv2d_bias" | |||
There was a problem hiding this comment.
dim parameter is not used at this moment.. Are you going to create separate variable names depending on dims? Isn't it?
|
So @dosier if you have any questions, ask me. I tried to comment on the existing solution and answer a few questions (and made more:) Great to find any existing NN architecture which uses this layer to attach in comments |
Early works for issue:
#125