11. Additional features

Here we describe some features covered by ai-project-templates that could be useful for your project. Keep in mind that these templates aren't aimed for model logic manipulation and only offer support for the prescribed models and the ones included in openmmlab repositories, the next features are provided without warranty or support.

Tensor output extraction

Sometimes you need to capture somehow the output from a specific layer, this sounds like an easy task but can get very difficult quickly if you don't have direct access to the model code. Hopefully all the pytorch models define an inner framework which allows to extract data from any layer of a model. Extract net tensor script is implemented for this task, given a model, the name of the specific layer the output is desired and and output path, will calcule the output of that specific layer and save it in the previosly specified output path.