Convolutional architectures in maDLC2.2


I am wondering to know which CNN models are available for MaDLC 2.2? As I found in this article ¨Multi-animal pose estimationand tracking with DeepLabCut¨, ImageNet-pretrained ResNets (22),the current state-of-the art model on the ImageNet bench-mark, EfficientNets (23), and introduce a novel multi-scale architecture (DLCRNet_ms5) we developed, which isloosely inspired by HRNet .

BUt when I run the tool, it shows

note to user: currently only ResNet50-v1 is available for maDLC
note to user: currently model comparision is not available in maDLC

Please let me know how we can use other CNN architectures

Hello! Make sure you upgrade deeplabcut to the newest 2.2rc1; the note to user will be gone, and you’ll have access to the new models from the GUI :slight_smile:

Thanks for your reply Jessy! I became a little confuse. In the newest version (2.2rc1), you proposed a multi-fusion architecture such as Figure1 in this article ¨Multi-animal pose estimation and tracking with DeepLabCut¨. But the old version of MaDLC is based on ResNet and deconvolutional layers. Right?