jxtps

The JavaCPP author states that the C API is included in the JavaCPP Presets for Tensorflow.

Would it make sense for TF4S to use [JavaCPP Presets for TensorFlow](https://github.com/bytedeco/javacpp-presets/tree/master/tensorflow)? They have convenient packaging of all the binaries for several platforms (linux, mac, windows), and GPU support is...

I guess this would kind of mimic how the D optimization uses both the fake & the real data to have the gradients cancel out where they're the same (...

It's been more than 2 years and this is still an issue. From what I can tell this is a textbook daemon thread? Can we please do at least one...

It appears that Android doesn't really care if a thread is marked as a daemon or not: https://stackoverflow.com/a/15323954/708381 > For Android purposes, daemon threads or non-daemon threads are moot. Your...

User threads block the VM from exiting. Daemon threads do not. That is their superficial difference. According to https://stackoverflow.com/a/2213443/708381 (and this matches what I've read elsewhere), daemon threads are abandoned...

If you're still concerned about Android (and I don't blame you if you are), then how about: ```java if (!"The Android Project".equals(System.getProperty("java.vendor"))) sThread.setDaemon(true); ``` Based on https://stackoverflow.com/questions/4519556/how-to-determine-if-my-app-is-running-on-android and https://developer.android.com/reference/java/lang/System#getProperties()

@dreiss can we please get this resolved? I'm happy to contribute whichever of the lines below you'll accept: ```java if (!"The Android Project".equals(System.getProperty("java.vendor"))) sThread.setDaemon(true); if ("true".equals(System.getProperty("com.facebook.fbjni.setDaemon"))) sThread.setDaemon(true); if ("true".equals(System.getProperty("fbjni.setDaemon"))) sThread.setDaemon(true);...

Don't get me wrong - the Stripe dashboard log is outstanding and we've used it several times. The reason we need the logging "in-house" is 1. what if the call...

The root of the resolution difference seems to be https://github.com/qubvel/segmentation_models.pytorch/blob/master/segmentation_models_pytorch/decoders/fpn/decoder.py#L86 ```python self.p5 = nn.Conv2d(encoder_channels[0], pyramid_channels, kernel_size=1) self.p4 = FPNBlock(pyramid_channels, encoder_channels[1]) self.p3 = FPNBlock(pyramid_channels, encoder_channels[2]) self.p2 = FPNBlock(pyramid_channels, encoder_channels[3]) ``` In...