我已经在BMXnet中使用resnet-binary模型来训练自己的数据,但是得到的是float权重而不是二进制权重。为什么?
网络参数
加载经过训练的模型并使用# -*- coding: utf-8 -*-
import concurrent.futures
import requests
import queue
import threading
# URL Pool
URLS = [
# Put your urls here
]
# Time interval (in seconds)
INTERVAL = 5 * 60
# The number of worker threads
MAX_WORKERS = 4
# You should set up request headers
# if you want to better evade anti-spider programs
HEADERS = {
'Accept': '*/*',
'Accept-Encoding': 'gzip, deflate',
'Accept-Language': 'en-US,en;q=0.9',
'Cache-Control': 'max-age=0',
'Connection': 'keep-alive',
#'Host': None,
'If-Modified-Since': '0',
#'Referer': None,
'User-Agent': 'Mozilla/5.0 (Windows NT 6.1) AppleWebKit/537.36 (KHTML, like Gecko) Chrome/67.0.3396.62 Safari/537.36',
}
############################
def handle_response(response):
# TODO implement your logics here !!!
raise RuntimeError('Please implement function `handle_response`!')
# Retrieve a single page and report the URL and contents
def load_url(session, url):
#print('load_url(session, url={})'.format(url))
response = session.get(url)
if response.status_code == 200:
# You can refactor this part and
# make it run in another thread
# devoted to handling local IO tasks,
# to reduce the burden of Net IO worker threads
return handle_response(response)
def ThreadPoolExecutor():
return concurrent.futures.ThreadPoolExecutor(max_workers=MAX_WORKERS)
# Generate a session object
def Session():
session = requests.Session()
session.headers.update(HEADERS)
return session
# We can use a with statement to ensure threads are cleaned up promptly
with ThreadPoolExecutor() as executor, Session() as session:
if not URLS:
raise RuntimeError('Please fill in the array `URLS` to start probing!')
tasks = queue.Queue()
for url in URLS:
tasks.put_nowait(url)
def wind_up(url):
#print('wind_up(url={})'.format(url))
tasks.put(url)
while True:
url = tasks.get()
# Work
executor.submit(load_url, session, url)
threading.Timer(interval=INTERVAL, function=wind_up, args=(url,)).start()
答案 0 :(得分:0)
似乎需要对模型进行二值化的步骤。查看该项目的仓库:https://github.com/hpi-xnor/BMXNet
通过在该存储库的“问题”标签中发布,您可能会获得更多的专业帮助。