If all permute layers (E, F) are commented out and only one transform layer (B) or (C) or (D) is commented in and the outputs of Model is set to, the model also compiles and the TocoConverter produces a tflite-file.īut for each combination of (B-C) and (E, F) being commented in so that one transform layer and one permute layer is there, the TocoConverter doesn't produce a tflite-file although the model compiles and the model summary looks as it should. If all transform layers (A-D) are commented out and only one permute layer (E) or (F) is commented in, directly gets passed input and an input shape of (2,3,1) for (E) or (0,2,3,1) for (F), the model compiles and the TocoConverter produces a tflite-file. In the same way it doesn't matter whether the Keras permute layer (E) or the backend version (F) is commented in. However, lines (B), (C) and (D) all work fine and it doesn't matter which one is commented in. Its output shape would be (None, None, 10, 42). The native Keras reshape layer (A) is the only one that doesn't work because it doesn't support dimensionality reduction. The input layer expects the last dimension to be 1, the transform layer being either a reshape or a squeeze layer gets rid of the 1, and the permutation layer just pushes the first dimension after the batch size to the end. The code example below containing a model with an input, "transformation" and permute layer compiles fine on pc and it is also possible to call predict()on it and getting the expected results. Exact command to reproduce: see minimum example code below.GPU model and memory: NVIDIA TITAN Xp, 12196MiB.GCC/Compiler version (if compiling from source):.Bazel version (if compiling from source):.TensorFlow version (use command below): v1.10.1-0-g4dcfddc5d1 1.10.1.TensorFlow installed from (source or binary): binary.iPhone 8, Pixel 2, Samsung Galaxy) if the issue happens on mobile device:. OS Platform and Distribution (e.g., Linux Ubuntu 16.04): Linux Ubuntu 16.04.Have I written custom code (as opposed to using a stock example script provided in TensorFlow): No.The tensor are actually next to each other in memory. It also applies toĪs I understand, contiguous in PyTorch means if the neighboring elements in Which discusses the meaning of contiguous in Numpy. Unlike view(), the returned tensor mayīe not contiguous any more. transpose() can operate both onĬontiguous and non-contiguous tensor. One difference is that view() can only operate on contiguous tensor and the The resulting out tensor shares it’s underlying storage with the input tensor, so changing the content of one would change the content of the other. The given dimensions dim0 and dim1 are swapped. Returns a tensor that is a transposed version of input. Transpose(), like view() can also be used to change the shape of a tensorĪnd it also returns a new tensor sharing the data with the original tensor:
You will find that theirĭata pointers are the same. It turns out that to find theĭata pointer, we have to use the data_ptr() method. PyTorch repo and got answers from the developer.
That their underlying data the same? Why this difference? You see that id of a.storage() and b.storage() is not the same. When you print the id of original tensor and viewing tensor: The semantics of reshape() are that it may or may not share the storage and you don’t know beforehand.Īs a side note, I found that torch version 0.4.1 and 1.0.1 behaves differently If you need a copy use clone() if you need the same storage use view(). You can not count on that to return a view or a copy. It means that torch.reshape may return a copy or a view of the original Contiguous inputs and inputs with compatible strides can be reshaped without copying, but you should not depend on the copying vs. When possible, the returned tensor will be a view of input. Returns a tensor with the same data and number of elements as input, but with the specified shape. On the other hand, it seems that reshape() has been introduced in version If you change the tensor value in the returned tensor, the corresponding value The returned tensor shares the underling data with the original tensor. view() vs reshape() and transpose() view() vs transpose()īoth view() and reshape() can be used to change the size or shape of PyTorch provides a lot of methods for the Tensor type.
0 kommentar(er)
