I added new bodyparts and trained again, but there's a bug "a mismatch between the current graph and the graph from the checkpoint."

Usage & Issues
#1

I download Full_Macaque weight from DeepLabCut Model Zoo, and I labeled 100 frames of the video that I took. I use both of them trained my monkey very well.
Then, I added two additional bodyparts for my animal. However, when I trained again, the bug came out.
The bug showed: “This is most likely due to a mismatch between the current graph and the graph from the checkpoint. Please ensure that you have not altered the graph expected based on the checkpoint. Original error:”

0303_1
0303_11842×989 89.6 KB

0303_3
0303_31760×999 188 KB

0303_4
0303_41853×1001 184 KB

0303_5
0303_51850×299 57.6 KB

(Before I train, I have changed the “init_weights” that I trained before for both “test/pose_cfg” and “train/pose_cfg.yaml” file)

#2

As stated in dlc.train_network docs, there is a parameter called keepdeconvweights, after you set it to False the training will work.

1 Like
#3

Hi Konrad,
I typed this code and it worked very well. Thank you so much!