PR #47726: updated convolutional_recurrent.py to include 1D, 2D, and 3D ConvLSTM instances.
Imported from GitHub PR https://github.com/tensorflow/tensorflow/pull/47726 Updating ConvLSTM to include 1D & 3D cases, as suggested by [this thread](https://github.com/tensorflow/tensorflow/issues/42399). - Implementation involves creating Base N-dimensional `ConvRNN`, `ConvLSTMCell`, and `ConvLSTM` classes which are inherited from by the concrete implementations in `ConvLSTM1D`, `ConvLSTM2D`, and `ConvLSTM3D`. Currently limited to 3D as maximum Convolution dimensionality as that is the highest rank implemented in `keras.backend`. - Rewrote tests to include 1D & 3D cases. The error message present in ConvLSTM2D as documented by [this thread](https://github.com/tensorflow/tensorflow/issues/36901) is still persistent. Copybara import of the project: -- b2b92b83478dd2a821af9a83a65077e99fe5300d by Matt Lyon <matthewlyon18@gmail.com>: updated ConvLSTM -- 960eb402f8c2d4c1f5b14703824c0307ed8bc000 by Matt Lyon <matthewlyon18@gmail.com>: updates to coding style -- 660d02ebccbd1327d2674ff35bc0af2303c612d5 by Matt Lyon <matthewlyon18@gmail.com>: fix docstrings PiperOrigin-RevId: 379750075
Showing
想要评论请 注册 或 登录