I would like to know how pruning works in tlt specially in the context of object detection - detectnet/resnet10 and if it’s possible to use the tlt-prune tool to prune our own custom models ?
Pruning removes parameters from the model. See chapter 9 for more info.
As long as your custom model is tlt format file, pruning can work via tlt-prune.
Can you shed more light on tlt format file and conversion of other format models to tlt format.I would like to use tlt-prune and other tlt based functionalities on custom models.
The tlt format file means the file with extension tlt. TLT format file is only compatible in TLT. And “tlt-prune” only works in TLT.
What do you mean by “custom models”?
By custom models, I mean our own models trained on other frameworks and saved in formats other than tlt - like onnx or mxnet models. Is there a way to convert those format models to tlt and use tlt based functionalities like tlt-prune.
Sorry, this is not supported.