Can DetectNet and ImageNet be trained to detect objects?

Hello all. I am very new to using the Jetson TX2 and have been following the Two Day Developer tutorial. I have played around with ImageNet and DetectNet and it can clearly identify objects that are pre-trained (people, dogs, airplanes, etc).

I’ve read that DIGITS can be trained and used to recognize more specific objects. I, however, do not have the available 100GB of storage to download this software and just want to train DetectNet to recognize one kind of object. Is this possible? If so, how do I train it?

Thanks in advance!

Hi mwehrenberg, see this portion of the tutorial for training DetectNet:

https://github.com/dusty-nv/jetson-inference/blob/master/docs/detectnet-training.md#downloading-the-detection-dataset

It uses a 4.5GB subset of the MS-COCO dataset for the training, so you will not need 100GB storage for training DetectNet.

Hi dusty_nv. I actually have tried downloading DIGITS in the last couple hours, and my host PC is incapable of installing it because the GPU cannot handle Nvidia drive (my PC crashes every time I try and install a new version). Is there any alternative to train DetectNet without the use of DIGITS?

Thanks again for all your help.