Successfully run in Windows10 !

[WindVChen]

``As TASN uses "att_grid_generator_cuda" in its code, we should first run "python setup.py build/install".

However, in Windows10, we may encounter something wrong. And here is my solution, hope to be helpful. (It indeed takes my much time.)

First, if you use VS in Chinese, you need to first change line 299 (around) in "cpp_extension.py" like this:

 try:
        if sys.platform.startswith('linux'):
            minimum_required_version = MINIMUM_GCC_VERSION
            versionstr = subprocess.check_output([compiler, '-dumpfullversion', '-dumpversion'])
            version = versionstr.decode().strip().split('.')
        else:
            # minimum_required_version = MINIMUM_MSVC_VERSION
            # compiler_info = subprocess.check_output(compiler, stderr=subprocess.STDOUT)
            # match = re.search(r'(\d+)\.(\d+)\.(\d+)', compiler_info.decode().strip())
            # version = (0, 0, 0) if match is None else match.groups()
            print("________________windows operation system ______________________")
            minimum_required_version = MINIMUM_MSVC_VERSION
            compiler_info = subprocess.check_output(compiler, stderr=subprocess.STDOUT)
            print("________________compiler info:", compiler_info)
            # match = re.search(r'(\d+)\.(\d+)\.(\d+)', compiler_info.decode().strip())
            match = re.search(r'(\d+)\.(\d+)\.(\d+)', compiler_info.decode(' gbk').strip())
            version = (0, 0, 0) if match is None else match.groups()

The code commented out above is the original ones. (This step is to avoid the format problem: utf-8 and gbk.)

Next, we should change the codes in "tasn-pytorch\src".

For "att_grid_generator_cuda.cpp", you should change all "AT_CHECK" into "TORCH_CHECK".

And for "att_grid_generator_cuda_kernel.cu", you should remove all the "const" of the parameters in fuction "attgridgen_gpu". Just like this:

void attgridgen_gpu(at::Tensor attx, at::Tensor atty,
    at::Tensor map_xi, at::Tensor map_yi,
    at::Tensor index_x, at::Tensor index_y,
    int batch_size, int att_size, int out_size, 
    float threshold, int iters)
{...}

Then, the call "python setup.py build" and "python setup.py install" can run successfully.

[Tracy-git]

thk so much