search

DeepLabCut / napari-deeplabcut

a napari plugin for labeling and refining keypoint data within DeepLabCut projects
GNU Lesser General Public License v3.0
52 stars 22 forks source link

Incorrect comparison of images and the final markup file in h5 format #63

Closed Tanya16181618 closed 1 year ago

Tanya16181618 commented 1 year ago

Discussed in https://github.com/DeepLabCut/napari-deeplabcut/discussions/61

Originally posted by **Tanya16181618** February 9, 2023 Hi all! I am using dlc version 2.3.0. I came across the fact that in napari I have an incorrect match between the name of the photo and the markup itself in the h5 and csv file. Because of this, in my opinion, the neural network does not learn from me at all and I get bad results. Attached is an example of such markup from the labled_data folder. For example, for photo 10.png there corresponds line 13 and photo 18.png. I calculated it so the results began to deteriorate, I began to check the markup. Moreover, in the graphical editor, the points are marked correctly, but not in the file. Tell me what can be done about it? [3_Problem.zip](https://github.com/DeepLabCut/napari-deeplabcut/files/10698557/3_Problem.zip)
jeylau commented 1 year ago

Hello @Tanya16181618, this problem was fixed with https://github.com/DeepLabCut/napari-deeplabcut/commit/23bbb6fe80c01c9a66a168f6db8e37a85865ae0b. This was due to the way napari sorted images when their names were not zero-padded; i.e., photo 1.png should have been named 01.png, 2.png 02.png, etc. This would work now though with napari-deeplabcut==0.1.2.