algorithm structure

[D19CrMei]

What is the detailed algorithm structure of SLEAP network? For example, what does its activation function, backbone network, and convolutional layers look like? Where can I find out?

[roomrys]

Hi @D19CrMei,

SLEAP allows the option to use a few different networks such as our own legacy leap network, unet (the default), hourglass, resnet, and pretrained_encoder (which uses a unet with a pretrained encoder). This documentation is currently listed in our Developer API page on the SLEAP website.

Im: All the types of backbone networks SLEAP supports. Taken from the Model Configuration in the SLEAP GUI.

We could and should improve upon the descriptions of the networks with a few diagrams for assistance - so thank you for bringing this to our attention. For further reading on the technical details of SLEAP, we released a paper (SLEAP: Multi-animal pose tracking) last month which has a Network architectures subsection located near the end of the paper in Methods.

Let me know if this is what you were looking for.

Thanks,
Liezl

[D19CrMei]

I browsed the latest papers, but still couldn't find the specific composition of each module in sleep, such as the number of neural network layers and training methods. Also, I want to train my own dataset, any suggestions for building the dataset? | | 梅Dr.CrT | | ***@***.*** | 签名由网易邮箱大师定制 On 4/23/2022 00:22，Liezl ***@***.***> wrote： Hi @D19CrMei, SLEAP allows the option to use a few different networks such as our own legacy leap network, unet (the default), hourglass, resnet, and pretrained_encoder (which uses a unet with a pretrained encoder). This documentation is listed in our currently listed in our Developer API page on the SLEAP website. Im: All the types of backbone networks SLEAP supports. Taken from the Model Configuration in the SLEAP GUI. We could and should improve upon the descriptions of the networks with a few diagrams for assistance - so thank you for bringing this to our attention. For further reading on the technical details of SLEAP, we released a paper (SLEAP:Multi-animal pose tracking) last month which has a Network architectures subsection located near the end of the paper in Methods. Let me know if this is what you were looking for. Thanks, Liezl — Reply to this email directly, view it on GitHub, or unsubscribe. You are receiving this because you were mentioned.Message ID: ***@***.***>

[roomrys]

@D19CrMei we have a good tutorial on the website which should help you get going.