megvii-model/labelenc

labelenc's Issues

RuntimeError: Error compiling objects for extension when run ' python setup.py build develop --no-deps '

Traceback (most recent call last):
File "setup.py", line 66, in
setup(
File "/home/qian/anaconda3/envs/LabelEnc/lib/python3.8/site-packages/setuptools/init.py", line 153, in setup
return distutils.core.setup(**attrs)
File "/home/qian/anaconda3/envs/LabelEnc/lib/python3.8/distutils/core.py", line 148, in setup
dist.run_commands()
File "/home/qian/anaconda3/envs/LabelEnc/lib/python3.8/distutils/dist.py", line 966, in run_commands
self.run_command(cmd)
File "/home/qian/anaconda3/envs/LabelEnc/lib/python3.8/distutils/dist.py", line 985, in run_command
cmd_obj.run()
File "/home/qian/anaconda3/envs/LabelEnc/lib/python3.8/distutils/command/build.py", line 135, in run
self.run_command(cmd_name)
File "/home/qian/anaconda3/envs/LabelEnc/lib/python3.8/distutils/cmd.py", line 313, in run_command
self.distribution.run_command(command)
File "/home/qian/anaconda3/envs/LabelEnc/lib/python3.8/distutils/dist.py", line 985, in run_command
cmd_obj.run()
File "/home/qian/anaconda3/envs/LabelEnc/lib/python3.8/site-packages/setuptools/command/build_ext.py", line 79, in run
_build_ext.run(self)
File "/home/qian/anaconda3/envs/LabelEnc/lib/python3.8/site-packages/Cython/Distutils/old_build_ext.py", line 186, in run
_build_ext.build_ext.run(self)
File "/home/qian/anaconda3/envs/LabelEnc/lib/python3.8/distutils/command/build_ext.py", line 340, in run
self.build_extensions()
File "/home/qian/anaconda3/envs/LabelEnc/lib/python3.8/site-packages/torch/utils/cpp_extension.py", line 708, in build_extensions
build_ext.build_extensions(self)
File "/home/qian/anaconda3/envs/LabelEnc/lib/python3.8/site-packages/Cython/Distutils/old_build_ext.py", line 195, in build_extensions
_build_ext.build_ext.build_extensions(self)
File "/home/qian/anaconda3/envs/LabelEnc/lib/python3.8/distutils/command/build_ext.py", line 449, in build_extensions
self._build_extensions_serial()
File "/home/qian/anaconda3/envs/LabelEnc/lib/python3.8/distutils/command/build_ext.py", line 474, in _build_extensions_serial
self.build_extension(ext)
File "/home/qian/anaconda3/envs/LabelEnc/lib/python3.8/site-packages/setuptools/command/build_ext.py", line 196, in build_extension
_build_ext.build_extension(self, ext)
File "/home/qian/anaconda3/envs/LabelEnc/lib/python3.8/distutils/command/build_ext.py", line 528, in build_extension
objects = self.compiler.compile(sources,
File "/home/qian/anaconda3/envs/LabelEnc/lib/python3.8/site-packages/torch/utils/cpp_extension.py", line 529, in unix_wrap_ninja_compile
_write_ninja_file_and_compile_objects(
File "/home/qian/anaconda3/envs/LabelEnc/lib/python3.8/site-packages/torch/utils/cpp_extension.py", line 1354, in _write_ninja_file_and_compile_objects
_run_ninja_build(
File "/home/qian/anaconda3/envs/LabelEnc/lib/python3.8/site-packages/torch/utils/cpp_extension.py", line 1683, in _run_ninja_build
raise RuntimeError(message) from e
RuntimeError: Error compiling objects for extension
(LabelEnc) qian@debian:~/LabelEnc$

Questions about class LabelEncodingFunction(nn.Module): in label_encoding_function.py

Is it necessary to set 5 stages in this function? I mean if 3, 4, 6 stages can work

class LabelEncodingFunction(nn.Module):
def init(self, in_channels):
super(LabelEncodingFunction, self).init()
self.stage1 = nn.Conv2d(in_channels, 128, 7, 2, 3)
self.stage2 = Bottleneck(128, 64, 256, 1, False, 2, 1)
self.stage3 = nn.Sequential(
Bottleneck(256, 128, 512, 1, False, 2, 1),
Bottleneck(512, 128, 512, 1, False, 1, 1),
)
self.stage4 = nn.Sequential(
Bottleneck(512, 256, 1024, 1, False, 2, 1),
Bottleneck(1024, 256, 1024, 1, False, 1, 1),
)
self.stage5 = Bottleneck(1024, 512, 2048, 1, False, 2, 1)

def forward(self, x):
    x = self.stage1(x)
    outs = []
    for m in [self.stage2, self.stage3, self.stage4, self.stage5]:
        x = m(x)
        outs.append(x)
    return outs

Questions about the optimization process.

Hello, I'm wandering that the optimizaiton process between the experiment with and without the auxiliary loss. In step 1, Is there a big or small gap during optimazing the model.

Can this loss applied in YOLOv5?

Hi authors,

love your work. can you please share some hints on how to implement the loss to yolov5 backbone ?

Autoencoder loss of step1

I use the labelenc in a custom model. In training step1 the autoencoder loss are not going down。in my opinion，the autoencoder loss should be in sync with detection loss since the share the detection head. what's the reason of above pattern ?