Closed Aexolowski closed 11 months ago
You could try saving them as cvs files and then loading them with numpy?
import numpy as np
import tqdm
import os
csv_paths = [XXX]
coordinates = {}
for filepath in tqdm.tqdm(csv_paths):
# load data
data = np.loadtxt(filepath, delimiter=',', skiprows=1)
# reshape
num_timepoints = data.shape[0]
data = data.reshape(num_timepoints, -1, 2)
# add to dict
name = os.path.splitext(os.path.basename(filepath))[0]
coordinates[name] = data
Hi Caleb, thanks a lot, we will try that! Another question is how / if we can also use the confidence estimates from the DLC model?
Sure, you can just load them normal (without defining the coordinates)
_, confidences, bodyparts = kpms.load_keypoints(keypoint_data_path, 'deeplabcut')
The one thing to make sure of is that confidences
and coordinates
end up with the same keys. If they don't, then you'll either have to edit the filenames or modify the code I posted earlier to adjust the names as necessary.
Hi there! We are tracking with DLC, but are correcting the data post-hoc. The final format is an excel sheet with frames X labels (x coordinate, y coordinate). What is the best way to load these data for keypoint-moseq? Thanks!