Closed vladBaciu closed 6 months ago
The model zoo we provide is entirely optional. There are a few ways to use your own models instead
mlonmch flow ...
command.environment.yml
with any custom directories like this:
paths:
models:
- "/work/git/mlonmcu/models"
- "/home/ga87puy/.config/mlonmcu/models"
- "/path/to/model/zoo"
The only limitation is that your model type has to be supported by one of mlonmcus frontends.
Regarding datasets: Since mlonmcu is only working with pre-trained models, it does not really care about datasets. The only case when the data becomes relevant is when validation the model outputs against a golden reference. This feature is supported by MLonMCU (see https://github.com/tum-ei-eda/mlonmcu/blob/main/ipynb/Examples/ValidateOutputs.ipynb) but we are currently working on a more sophisticated solution for model output verification...
I hope this answers your question!
Can custom models and datasets be used easily on this project ?
I am wondering if I can evaluate a custom model on a custom dataset that is not publicly available. If yes, could you give me some hints ?