Incorrect weights for TRAILMAP_MS ?

[C-Achard]

After downloading the TRAILMAP_MS weights for inference, I get the following :
Error(s) in loading state_dict for TRAILMAP_MS:
Missing key(s) in state_dict:
(long list of missing keys...)
Unexpected key(s) in state_dict:
(long list of unexpected keys...)

Could it be that they are not the correct ones ? How easily can we update the weights @ http://deeplabcut.rowland.harvard.edu/cellseg3dmodels ?

[MMathisLab]

these are the best mesoSPIM trained weights for this model, per @vidalmaxime

[C-Achard]

I just think they are paired to the wrong model, so it causes an error in the plugin.

TODO:

• Possibly just swap model links
• Rename models correctly.

TRAILMAP --> trailmap-pytorch
TRAILMAP_MS --> trailmap-ms trained on mesospim data

[MMathisLab]

yes, so to clarify: 😎

There are 4 models (+ unet, which isn't supported actually?):

and TRAILMAP --> trailmap-pytorch
TRAILMAP_MS --> trailmap-ms trained on mesospim data

[C-Achard]

@MMathisLab The Unet folder is for what is currently called TRAILMAP_MS, it's simply code from https://github.com/wolny/pytorch-3dunet Maxime used to re-create TRAILMAP in a more robust way in Pytorch

[C-Achard]

Naming issues have been addressed in #19, see 34da28b.

However, the weights file for TRAILMAP_PyTorch on the server is still incorrect, I can provide one when appropriate so we can update it on the server.