安装、使用 tensorflow 遇到的问题

 Q1: 没发现满足需求的版本

ERROR: Could not find a version that satisfies the requirement keras-nightly~=2.5.0.dev (from tensorflow) (from versions: none)
ERROR: No matching distribution found for keras-nightly~=2.5.0.dev

按照官方文档Installation — TensorFlow 2 Object Detection API tutorial documentation安装时总是出现错误。问题出在 keras-nightly. keras-nightly 是 Keras 的每日构建版本（开发版），通常包含未正式发布的实验性代码。TensorFlow 2.5.0 在开发阶段可能依赖此版本，但正式发布后不再需要。当前 PyPI 镜像源（如清华源）可能已移除陈旧的 keras-nightly 开发版，导致安装失败。

solutions：

绕过版本2.5.0，下载清华源中的较新版本。 

pip install tensorflow -i https://pypi.tuna.tsinghua.edu.cn/simple

Attention：如果在实际运行项目时，发现缺失了包，建议到同一个源进行下载，这样无需纠结是否适配问题，是保持一致性（keep consistence)的做法。其他源的包并不合适。

Q2:

很笨的问题，没有进入python环境

import  torch失败、import-im6.q16: not authorized ‘torch’@ error/constitute.c/WriteImage/1037. 问题的解决_not authorized `torch-CSDN博客Q3：socket.gaierror: [Errno -2] Name or service not known

关于socket.gaierror: [Errno -2] Name or service not known报错解决-CSDN博客

对于 ROS 单机通信的环境配置，需要确保 ~/.bashrc 文件中写入

export ROS_HOSTNAME=localhost
export ROS_MASTER_URI=http://localhost:11311

以及在 /etc/hosts 文件中存在

127.0.0.1 localhost

并不存在多余的其他信息，如

127.0.0.1 ubuntu

否则系统不清楚究竟该使用哪个。我就是这里报错。

Q4: 在自己运行脚本时报错，下面是一大段log。

Traceback (most recent call last):
  File "/home/ncut/miniconda3/envs/tf/lib/python3.9/tarfile.py", line 191, in nti
    n = int(s.strip() or "0", 8)
ValueError: invalid literal for int() with base 8: 'ts" cont'

During handling of the above exception, another exception occurred:

Traceback (most recent call last):
  File "/home/ncut/miniconda3/envs/tf/lib/python3.9/tarfile.py", line 2617, in next
    tarinfo = self.tarinfo.fromtarfile(self)
  File "/home/ncut/miniconda3/envs/tf/lib/python3.9/tarfile.py", line 1295, in fromtarfile
    obj = cls.frombuf(buf, tarfile.encoding, tarfile.errors)
  File "/home/ncut/miniconda3/envs/tf/lib/python3.9/tarfile.py", line 1237, in frombuf
    chksum = nti(buf[148:156])
  File "/home/ncut/miniconda3/envs/tf/lib/python3.9/tarfile.py", line 193, in nti
    raise InvalidHeaderError("invalid header")
tarfile.InvalidHeaderError: invalid header

During handling of the above exception, another exception occurred:

Traceback (most recent call last):
  File "/home/ncut/miniconda3/envs/tf/lib/python3.9/site-packages/tensorflow_hub/resolver.py", line 192, in download_and_uncompress
    file_utils.extract_tarfile_to_destination(
  File "/home/ncut/miniconda3/envs/tf/lib/python3.9/site-packages/tensorflow_hub/file_utils.py", line 47, in extract_tarfile_to_destination
    with tarfile.open(mode="r|*", fileobj=fileobj) as tgz:
  File "/home/ncut/miniconda3/envs/tf/lib/python3.9/tarfile.py", line 1851, in open
    t = cls(name, filemode, stream, **kwargs)
  File "/home/ncut/miniconda3/envs/tf/lib/python3.9/tarfile.py", line 1732, in __init__
    self.firstmember = self.next()
  File "/home/ncut/miniconda3/envs/tf/lib/python3.9/tarfile.py", line 2629, in next
    raise ReadError(str(e))
tarfile.ReadError: invalid header

During handling of the above exception, another exception occurred:

Traceback (most recent call last):
  File "/home/ncut/code_garden/face_demo1.py", line 7, in <module>
    model = hub.load("https://tfhub.dev/tensorflow/blazeface/1")
  File "/home/ncut/miniconda3/envs/tf/lib/python3.9/site-packages/tensorflow_hub/module_v2.py", line 100, in load
    module_path = resolve(handle)
  File "/home/ncut/miniconda3/envs/tf/lib/python3.9/site-packages/tensorflow_hub/module_v2.py", line 55, in resolve
    return registry.resolver(handle)
  File "/home/ncut/miniconda3/envs/tf/lib/python3.9/site-packages/tensorflow_hub/registry.py", line 49, in __call__
    return impl(*args, **kwargs)
  File "/home/ncut/miniconda3/envs/tf/lib/python3.9/site-packages/tensorflow_hub/compressed_module_resolver.py", line 81, in __call__
    return resolver.atomic_download(handle, download, module_dir,
  File "/home/ncut/miniconda3/envs/tf/lib/python3.9/site-packages/tensorflow_hub/resolver.py", line 421, in atomic_download
    download_fn(handle, tmp_dir)
  File "/home/ncut/miniconda3/envs/tf/lib/python3.9/site-packages/tensorflow_hub/compressed_module_resolver.py", line 78, in download
    return resolver.DownloadManager(handle).download_and_uncompress(
  File "/home/ncut/miniconda3/envs/tf/lib/python3.9/site-packages/tensorflow_hub/resolver.py", line 200, in download_and_uncompress
    raise IOError("%s does not appear to be a valid module." % self._url)
OSError: https://tfhub.dev/tensorflow/blazeface/1 does not appear to be a valid module.

前几次碰见，慌张得很，认为是 我 pip -install 出现了问题。实则不然。这段代码的真正问题在

import cv2
import tensorflow as tf
import tensorflow_hub as hub
import numpy as np

model = hub.load("https://tfhub.dev/tensorflow/blazeface/1")

它的原因是下载的包文件损坏，无法通过该方式进行下载。有两种情况，

• 网络不稳定导致下载的包（通常是 .tar.gz 或 .whl）不完整或损坏。

• 服务器上的包文件本身存在问题（罕见但可能）。

此处是该路径根本无法正确解析。

我在碰到这个报错时，很大的误区是担心 pip 下载的包有问题，实际上问题并不在那里，在 hub.load()提供的字符串不正确。这种结论可以通过试验不同的python脚本得出。