Closed mrnels19 closed 5 months ago
I'm having the same issues - training on colab (sleap version 1.3.3) isn't generating metric files for my top-down models either. When I was running training on my own machine (sleap version 1.3.2) they were generated. One difference is that I manually stopped training on colab, while it ran until plateauing on my own machine.
I have trained a SLEAP model with a single instance using Google Colab. However, training with a single instance does not give me the option to view metrics, such as the error distance, localization error, and object keypoint similarity.
Since the metrics of error distance, localization error, and object keypoint similarity are used to assess the accuracy and validity of models with multiple instances, then I think it would be helpful to include these metrics for models with a single instance. This is the training command for a single instance SLEAP model on Google Colab:
!sleap-train single_instance.json filename.pkg.slp
.I thought that I would be able to generated the metrics for a single instance SLEAP model, just as I can for a SLEAP model with multiple instances.
I am unable to load the metrics, as the single instance SLEAP model does not contain the files
metrics.train.npz
andmetrics.val.npz
.I am using SLEAP version 1.3.1 in Google Colab Pro +.
I have included the list of commands I have run in my Google Colab script, which has been adapted from the Google Colab examples on the SLEAP website. I experience an error when I load the last command in this list, which involves a metrics file that does not appear to be generated. I would like SLEAP to generate metrics files for single instance SLEAP models so that I can assess them. Or, can you show me a way to generate figures or other data that can showcase the accuracy and validity of a single instance SLEAP model?
i``` mport os os.chdir('/content/drive/My Drive/1.28.24_three_chamber')
from google.colab import drive drive.mount('/content/drive')
%load_ext tensorboard %tensorboard --logdir "/content/drive/My Drive/1.28.24_three_chamber" --reload_multifile=true --reload_interval=1 --max_reload_threads=2
import sleap import numpy as np import pandas as pd import matplotlib as mpl import matplotlib.pyplot as plt import seaborn as sns
mpl.style.use("seaborn-deep") sleap.versions()
metrics = sleap.load_metrics("models/240128_195816.single_instance", split="val") print("\n".join(metrics.keys()))