Request for the Continue training. And Updated version have dependency error,

[silverline-art]

Greetings,
The Problem: Currently there's no option to continue training of the already trained model or fine tuning the model with the updated dataset.
Please add this feature in next update please. For the Pose application.

Dependency Issue:
For the Updated version of the app

Installing collected packages: lightning
Attempting uninstall: lightning
Found existing installation: lightning 2.2.5
Uninstalling lightning-2.2.5:
Successfully uninstalled lightning-2.2.5
ERROR: pip's dependency resolver does not currently take into account all the packages that are installed. This behaviour is the source of the following dependency conflicts.
lightning-pose-app 1.5.0 requires lightning[app]==2.2.5, but you have lightning 2.5.1.post0 which is incompatible.
Successfully installed lightning-2.5.1.post0
/home/shivam/Desktop/Python_Venvs/lia/bin/python3.10: No module named lightning.main; 'lightning' is a package and cannot be directly executed

[themattinthehatt]

Hi @silverline-art thanks for the issue.

With regards to updating the dataset/adding more labels, we've found that it's usually better to just train the model from scratch using the full dataset, rather than finetuning on the labels.

For continuing model training, can you let me know what your use-case is? Is it to fine-tune an already trained model, or do you, for example, have a model you trained for 300 epochs and now you want to train it for an additional 200 epochs?

For the dependency issue, please make sure to not upgrade your lightning version whenever you launch the app. You'll need to uninstall your current version of lightning and reinstall the old version for the app to work; please see the FAQ "I accidentally upgraded my lightning version, what should I do?" here: https://pose-app.readthedocs.io/en/latest/source/faqs.html

[silverline-art]

Thank you for your response:
For continuing model training; It is for fine-tune you can say I think.
2 cases:
Case 1: As I usually try to test both Semi-/Supervised Model with different numbers of frames in datasets, like 20 frames, 100 frames 300 frames.
Case 2: I'm also interested training the existing model with new dataset of similar in nature. for the purpose of fine-tuning and making the model more robust.