knuckles | Knuckles finger tracking and advanced interaction system | Plugin library

Usually, neural networks will have a very hard time to predict exact coordinates of landmarks. A better approach is probably a fully convolutional network. This would work as follows:

1. You omit the dense layers at the end and thus end up with an output of (m, n, n_filters) with m and n being the dimensions of your downsampled feature maps (since you use maxpooling at some earlier stage in the network they will be lower resolution than your input image).
2. You set n_filters for the last (output-)layer to the number of different landmarks you want to detect plus one more to indicate no landmark.
3. You remove some of the max pooling such that your final output has a fairly high resolution (so the earlier referenced m and n are bigger). Now your output has shape mxnx(n_landmarks+1) and each of the nxm (n_landmark+1)-dimensional vectors indicate which landmark is present as the position in the image that corresponds to the position in the mxn grid. So the activation for your last output convolutional layer needs to be a softmax to represent probabilities.
4. Now you can train your network to predict the landmarks locally without having to use dense layers.

This is a very simple architecture and for optimal results a more sophisticated architecture might be needed, but I think this should give you a first idea of a better approach than using the dense layers for the prediction.

And for the explanation why your network does predict the same values every time: This is probably, because your network is just not able to learn what you want it to learn because it is not suited to do so. If this is the case, the network will just learn to predict a value, that is fairly good for most of the images (so basically the "average" position of each landmark for all of your images).

Before pull queries were introduced, ksqlDB nodes did not talk to each other. The 'clustering' of ksqlNodes leveraged a shared command topic and the Kafka consumer groups protocol of Kafka to share out work. This is why you're not seeing any messages about trying to join the cluster: the node doesn't join the cluster.

Pull queries using Kafka Streams' interactive queries under the hood. This works by each ksqlDB node advertising an endpoint other nodes can reach it. These advertised endpoints are shared between ksqlDB nodes using the Kafka consumer group protocol, i.e. the nodes become aware of each other through their communication with Kafka.

When you issue a pull query to a node that doesn't host the information you require it will attempt to forward the request to the node that does by contacting it on the endpoint it advertised. In your case, what is happening is that the endpoint your nodes from one machine are advertising is not accessible from the other machine.

KsqlDB v0.8 - 0.9 takes the first listener in the listeners config and uses this as its advertised listener / endpoint. A common problem is people use a wildcard or loopback address in their listener config, e.g.