SthPhoenix

Hi! I have used exactly the same script from insightface repo.

Hi! Yes, I have used this config, I have trained models on single GPU, some on rtx2080, some on rtx3090. I can't recall if I have changed batch size, but...

Hi! Batch inference is already supported for all recognition models and for SCRFD and YOLOv5 family detection models.

Thanks! I'll give it a try at monday (left all my code at work), also it seems that for some reason hex values from **Thinbus** for some reason have wrong...

Update: looks like **thinbus** calculate M value as: M= H(A,B,K) While in **pysrp**: M=H(H(N) XOR H(g), H(I),s,A,B,K) So here **pysrp** is much closer to original specs than **thinbus**

Ok, I beat it. In short: - **pysrp** mostly calculates hashes from byte strings, while in **thinbus** all hashes are based on hex strings. - there are a bit different...

I have actually used the **5054** branch, but still had to change some calculations in order to made two libs communicate. I made all changes on server side, since I...

Still testing, got some issues with authentication from time to time. Approximately 3 out of 10 tries generates wrong keys. Seems that it's something connected to trimming leading zeroes on...

Got it fully working and stable, also added simple safe prime config generator, to simply export server side configs into thinbus client. I'll tidy up a code a little and...

Created fork here: [pysrp_thinbus](https://github.com/SthPhoenix/pysrp_thinbus)