about the network about ringnet HOT 3 CLOSED

1. The trick is in batching the data. yes the concept is similar like Siamese networks extended to rings. One can pass all the data through one encoder and slice the data according to their labels at the end of the encoder output to consider it for different rings and compute the loss.
2. Only the 100 shape related vectors are considered.

from ringnet.

I have a question during reading the paper. Did the R ring elements mean there is only one network(encoder), and for each step, use the same network(encoder) compute(encode) R images, then use the R results compute Lsc，just like Siamese network?
Another question is when compute the Lsc, did you use all 159-d output or only the 100 shape related.

from ringnet.

Releasing the training code is having some internal licensing issues due to its reliance in tensorflow flame which may take further time.

from ringnet.