YttriLab / A-SOID

An active learning platform for expert-guided, data efficient discovery of behavior.
Other
54 stars 8 forks source link

Error in predicting labels #98

Open honamkung opened 2 months ago

honamkung commented 2 months ago

Describe the bug I trained a pose estimation model in SLEAP, which was then uploaded into ASOiD along with an annotation file in CSV format. I was able to pass both the feature extraction and active learning steps without any errors. However, during the prediction stage, predicting new data with an uploaded config file resulted in an error. I have attached a screenshot of the error for your reference. Given that this error didn't occur with our previous 5-node skeleton design (as presented in your Nature Methods paper), I suspect it may be related to our new 9-node skeleton design. Is there a specific skeleton/node design that works with ASOiD? We recently changed the skeleton design to better train a SLEAP model, as a shallow tree-like design with a bottom-up approach is known to improve model training. Could you please review the attached screenshot of the error and our latest skeleton design?

Screenshot of the error Screenshot 2024-08-05 214524

Our skeleton design

thumbnail_image003
JensBlack commented 2 months ago

Can you please provide more information about your project as suggested in the standard issue template?

There is no requirement for a specific set of keypoints. So it should work, if the same keypoints are present in the pose file you predicted.

this might be related to #82 #73