Transfer Learning Toolkit and DIGITS

I want to improve a pre-trained model to detect people from above. What is the difference if I trained the model in TLT and DIGITS. From what I understand, both can improve pre-trained model and generate ready to deploy model for Jetson Nano.

Hi marco_savoia,
The detectnet_v2 network in TLT is similar to DIGITS.

What I mean is more to the what objective and purpose difference between TLT and DIGITS? Because DIGITS and TLT offer improve pre-trained model functionality.

Hi marco,
In TLT, ngc already provides the pre-trained models.
See https://docs.nvidia.com/metropolis/TLT/tlt-getting-started-guide/index.html#downloading_models

Hi. I need use my models (CNN’s) made in digits (caffe) and improve with TLT. It’s possible? If yes, How I do?

Other, anyone know How prune layer or connection in caffe with digits?
Thanks

Only tlt format models can be pruned.
NVIDIA GPU Cloud(NGC) provides the pretrained weights.
We cannot ensure other pre-trained weights can work.

Understand. Thanks.
One more question. There are problem docker digits-tensorflow for Driver Version: 440.82 and CUDA Version: 10.2?
Digits run but not work when I creat job for model in TensorFlow.
Thanks again.

Sorry, I did not understand what your problem is. Could you please give more details or the steps for reproducing? BTW, if it is not a TLT issue, I am afraid you can create a new topic under other forums.