High noise and causal kernel models for GCaMP8s 45Hz

[sumiya-kuroda]

Hello @PTRRupprecht ,

Hope you are doing well! We are happy with the Cascade models you made for GCaMP8s below, and we would like to have their high noise versions and causal kernel versions as well. Would you be able to train the models and add them to the repo for us, please?

• GC8s_EXC_45Hz_smoothing20ms:
• GC8s_EXC_45Hz_smoothing50ms:
• GC8s_EXC_45Hz_smoothing100ms:

Thanks,
Sumiya

[PTRRupprecht]

Hi @sumiya-kuroda,

Sure, this should not be a problem at all. I will initiate the training process and let you know once the models are available!

Best,
Peter

[PTRRupprecht]

@sumiya-kuroda

P.S. The current models (that you mentioned) go up to noise levels of 28. Therefore, although "high_noise" is not explicitly mentioned in the model name, they are actually "high-noise" compatible. Or do you need models trained for even higher noise levels? I have never seen data with noise levels >25, especially not when frame rates are high. I would therefore train the causal kernel versions with the same noise level ranges as the ones trained without causal kernel.

[sumiya-kuroda]

Hello @PTRRupprecht ,

Oh I see! I asked for the high noise model in case i get even noisier data from our future microprism recordings. But in that case I only need the causal kernel model. Thank you very much for clarification!!!

Best,
Sumiya

[PTRRupprecht]

Hi @sumiya-kuroda,

The new models are now online and available:

GC8s_EXC_45Hz_smoothing20ms_causalkernel_high_noise
GC8s_EXC_45Hz_smoothing50ms_causalkernel_high_noise
GC8s_EXC_45Hz_smoothing100ms_causalkernel_high_noise

If you find out that you indeed need models for even noisier data, just reach out, then we can fix it. From my experience, microprism data can be really noisy, but I don't expect them to be not covered by the available models.

And let me know if you run into any other problems!

Best,
Peter

[sumiya-kuroda]

Thank you very much! So fast! I will try those models later today! :)