denru01/netadapt

This repo contains the official Pytorch reimplementation of the paper "NetAdapt: Platform-Aware Neural Network Adaptation for Mobile Applications".

netadapt's Issues

No progress of algorithm

Hi!

I've tried running the code you've provided on MobileNet, the pretrained model, which was given by you as an example. I believe that calculating LUTs went correctly. Then I've tried applying the algorithm with 0,5 latency, by running:

sh scripts/netadapt_mobilenet-0.5latency.sh

but after some time the algorithm stopped proceeding on:

Launch a worker for block 13 ['/usr/local/bin/python', 'worker.py', 'models/mobilenet/prune-by-latency/worker', 'models/mobilenet/prune-by-latency/master/iter_0_best_model.pth.tar', '13', 'LATENCY', '0.033316052734851845', '1', '500', '0', 'latency_lut/lut_mobilenet.pkl', 'data/', '3', '224', '224', 'mobilenet', '0.001'] Update job list: [{'iteration': 1, 'block': 1, 'gpu': 1}, {'iteration': 1, 'block': 2, 'gpu': 2}, {'iteration': 1, 'block': 3, 'gpu': 3}, {'iteration': 1, 'block': 4, 'gpu': 4}, {'iteration': 1, 'block': 5, 'gpu': 5}, {'iteration': 1, 'block': 6, 'gpu': 6}, {'iteration': 1, 'block': 13, 'gpu': 0}] Update available gpu: []

Despite of 3 days of computing it didn't progress more. Could you advise me if something is wrong there?

Regards,
Piotrek

raise ValueError

raise ValueError('target_resource {} is not achievable in iter {}.'.format(target_resource, current_iter))
ValueError: target_resource 718242470.4 is not achievable in iter 1.

public netadaptv2

Hi,
Thanks for you excellent work.
When will you release netadaptv2?

Tensorflow model

Hi!

Is there a possibility that you provide tensorflow implementation of NetAdapt?
Or, is it possible to run that pytorch implementation on tensorflow saved_model or frozen graph?

Regards,
Piotrek

where is long-term fine-tune

Hi,thank you for your work.I trained the code in my project, and I can not find the long-term fine tune in the master.py.Can you help me?Thank you very much!