search

kahst / BirdNET-Analyzer

BirdNET analyzer for scientific audio data processing.
Other
859 stars 157 forks source link

GUI Train error when using append #385

Closed shnchh closed 1 month ago

shnchh commented 3 months ago

Describe the bug When using append model save mode, the trainer runs until completion, then produces the error screen along with a message "No such file or directory". No error message found in log other than a TensorFlow warning on deprecated cross entropy loss function. The GUI settings screen shows "no content" under error log file. The tflite file is output but there is no labels and parameters file.

This issue doesn't happen if I choose replace and not append.

To Reproduce Steps to reproduce the behavior:

  1. Go to Train
  2. Added two classes of .wav files.
  3. Save as tflite file, append mode, everything else default, start training.
  4. See error

Expected behavior No error

Desktop (please complete the following information):

Josef-Haupt commented 3 months ago

Sadly I can't reproduce this bug on my end, please post a screenshot of all the options used in the training tab.

shnchh commented 2 months ago

image image image

max-mauermann commented 2 months ago

I also cant reproduce this error in the current version with the parameters seen in the screenshots. Does it still occur on your system? If yes, would you mind sharing the data you use, this might also be something that causes the problem.

shnchh commented 2 months ago

I have currently organised my recordings into 3 folders: (I removed the bird species name as it is a sensitive species) image

Here is a sample recording from my "others" folder. Recording

I am suspecting the issue to be with the output folder, as I receive the "No such file or directory"error only after training is completed. Can I have more insight into what the append feature is doing? Does it still work if my species is already in the original birdnet species list? And does the output folder for the classifier have to contain a species list?

larskristian02 commented 1 month ago

I am experiencing the exact same issue and can't seem to find a solution. Have you managed to fix it?

shnchh commented 1 month ago

I did not, but I managed to work around it by working directly with the command line interface and fixing some minor bugs in the source code. e.g. python train.py --i "folder path" --o "outputfolder\\classifiername.tflite"

larskristian02 commented 1 month ago

Thank you! I'm new to python, so i'm just wondering: do you still use the BirdNET gui application, or do you do the training manually in python?

shnchh commented 1 month ago

Nope you would not use the GUI. You can follow the instructions found on the GitHub birdnet homepage, under Step 6. setup (depending on your OS) and 7.1 CLI