What's Changed * Update test badge by vuillaut in https://github.com/IndexedConv/IndexedConv/pull/33 * Create codemeta.json by vuillaut in https://github.com/IndexedConv/IndexedConv/pull/34 * Version 1.3.2 by vuillaut in https://github.com/IndexedConv/IndexedConv/pull/35
What's Changed * fix doc requirements file by vuillaut in https://github.com/IndexedConv/IndexedConv/pull/27 * Test action by vuillaut in https://github.com/IndexedConv/IndexedConv/pull/29 * Pypi autopublish by vuillaut in https://github.com/IndexedConv/IndexedConv/pull/30 * add sphinx-rtd-theme to doc requirements by vuillaut in https://github.com/IndexedConv/IndexedConv/pull/28 * complete doc requirements for RTD by vuillaut in https://github.com/IndexedConv/IndexedConv/pull/32 * Bump version to 1.3.1 by vuillaut in https://github.com/IndexedConv/IndexedConv/pull/31
* Update to PyTorch 1.4 * Replace tensorboardX by PyTorch Summarywriter
1.1
Convolution memory consumption the matmul function has been replaced by an explicit expand of the weight matrix and bmm leading to 30% less memory consumption. However, indexedconv still uses way more memory than torch.nn.Conv2d.