Add torch to build requirements

[r-b-g-b]

Since setup.py imports torch

YOLOX/setup.py

Line 8 in 6819425

import torch

installing yolox requires a separate install of torch beforehand (as documented in the README). However, there is a standard way to include build dependencies for a Python package. From the PyPA docs:

pyproject.toml tells build tools (like pip and build) what is required to build your project.

Adding pyproject.toml with a build-system.requires section that includes torch would eliminate the need to install torch with a separate command prior to installing yolox. (Importantly, this only installs torch during the build stage, it still needs to be installed into the Python environment to use the package). This makes it easier to use a library in other packages.

For example, the following fails:

pip install git+https://github.com/Megvii-BaseDetection/YOLOX.git

Collecting git+https://github.com/Megvii-BaseDetection/YOLOX.git
  Cloning https://github.com/Megvii-BaseDetection/YOLOX.git to /tmp/pip-req-build-wruzfplt
  Running command git clone -q https://github.com/Megvii-BaseDetection/YOLOX.git /tmp/pip-req-build-wruzfplt
  Resolved https://github.com/Megvii-BaseDetection/YOLOX.git to commit 6819425e78e28926d132863d4be20811729c9d44
    ERROR: Command errored out with exit status 1:
     command: /home/robert/anaconda3/envs/zamba-package/bin/python -c 'import io, os, sys, setuptools, tokenize; sys.argv[0] = '"'"'/tmp/pip-req-build-wruzfplt/setup.py'"'"'; __file__='"'"'/tmp/pip-req-build-wruzfplt/setup.py'"'"';f = getattr(tokenize, '"'"'open'"'"', open)(__file__) if os.path.exists(__file__) else io.StringIO('"'"'from setuptools import setup; setup()'"'"');code = f.read().replace('"'"'\r\n'"'"', '"'"'\n'"'"');f.close();exec(compile(code, __file__, '"'"'exec'"'"'))' egg_info --egg-base /tmp/pip-pip-egg-info-dveu39rs
         cwd: /tmp/pip-req-build-wruzfplt/
    Complete output (5 lines):
    Traceback (most recent call last):
      File "<string>", line 1, in <module>
      File "/tmp/pip-req-build-wruzfplt/setup.py", line 8, in <module>
        import torch
    ModuleNotFoundError: No module named 'torch'
    ----------------------------------------

Whereas this succeeds:

pip install git+https://github.com/r-b-g-b/YOLOX.git@build-requirements

Collecting git+https://github.com/r-b-g-b/YOLOX.git@build-requirements
  Cloning https://github.com/r-b-g-b/YOLOX.git (to revision build-requirements) to /tmp/pip-req-build-z748qonp
  Running command git clone -q https://github.com/r-b-g-b/YOLOX.git /tmp/pip-req-build-z748qonp
  Running command git checkout -b build-requirements --track origin/build-requirements
  Switched to a new branch 'build-requirements'
  Branch 'build-requirements' set up to track remote branch 'build-requirements' from 'origin'.
  Resolved https://github.com/r-b-g-b/YOLOX.git to commit 3631ad70d1b01d5988ff5745655eb2ff182d5bd9
  Installing build dependencies ... done
  Getting requirements to build wheel ... done
    Preparing wheel metadata ... done

[r-b-g-b]

@FateScript Any thoughts on this? I'd love to use YOLOX as a library in our package. Thanks!

[CLAassistant]

All committers have signed the CLA.

[FateScript]

Sorry for not not noticing this PR. LGTM. Thanks for contributing @r-b-g-b