Mask RCNN pruning problem

According to the previous topic,

it is necessary to retrain the model after pruning. But the size of the model increases again. So what is the benefit of pruning?

If the tlt model is pruned, the exported etlt model’s size will be smaller than unpruned one.
The fps(inference speed) is usually faster than unpruned one. That’s the benefit of pruning.
BTW, you can also check the trainable parameters in training/retraining log.

Yes, you are right. By pruning, the size of the model decreases. But according to this link, the mask rcnn pruned model must be retrained first and so, the model size increases again.

Need not care about the tlt model’s size, the tensorrt engine will be smaller than the unpruned one.

Thank you.
I will check

This topic was automatically closed 14 days after the last reply. New replies are no longer allowed.