File "c:\Users\Administrator\Desktop\LightNet-master\modules_ext_init_.py", line 2, in
from .__ext import lib as _lib, ffi as _ffi
ModuleNotFoundError: No module named 'modules._ext.__ext'

I can not find " .__ext" in your project
can you upload it?

@ansleliu , thanks for your nice work, but I am a little confused with your image pre-processing.


Which kind of ImageNet pretrained weights are you using for mobinenet, since the img pre-processing is different from those pre-weights provided by caffe and pytorch?

Hello, thanks for sharing your implementation. I did not understand why do you add 1 in the return of SCSE module

https://github.com/ansleliu/LightNet/blob/04af22501d644b30a93b30b28b32163d60ae2266/modules/misc.py#L110

I think it should be

return (chn_se + spa_se)

Basiclly, the ScSE did as follows: (1) convolution 1x1 x C for channel side and (2) convolution 1x1x1 for spatial size. Then add (1) and (2)

Just for enchancement, this repo are helpful for build realtime segmentation applications, but now that Pytorch official has been depracted cffi, would that be a good idea to upgrade into CppExtenstion under modules cuda and cpp files?

I feel very confused. I use the train_mobile.py to train the model with out pre-trained on the cityscapes dataset. I can only get the mIoU of 30~40. How can we get the mIoU of 70? How many epochs ？

Hi,

in modules/_ext/init.py, line 2
from .__ext import lib as _lib, ffi as _ffi

I couldn't find __ext, is there something missing?

I have installed inplace_abn.

Thanks for your help

I got some errors like:

RuntimeError: cuda runtime error (2) : out of memory

How to solve this problem?

Thanks

warnings.warn(ABI_INCOMPATIBILITY_WARNING.format(compiler))

+++++++++++++++++++++++++++++++++++++++++++++++++++++++

0. Setting up DataLoader...

+++++++++++++++++++++++++++++++++++++++++++++++++++++++

Found 8925 train images...
Found 1500 val images...

+++++++++++++++++++++++++++++++++++++++++++++++++++++++

1. Setting up Model...
   THCudaCheck FAIL file=/pytorch/aten/src/THC/THCGeneral.cpp line=74 error=30 : unknown error
   Traceback (most recent call last):
   File "train_mobile.py", line 374, in
   train(train_args, data_path, save_path)
   File "train_mobile.py", line 77, in train
   model = torch.nn.DataParallel(model, device_ids=[0,1]).cuda()
   File "/home/dl/anaconda2/envs/test_py3/lib/python3.6/site-packages/torch/nn/modules/module.py", line 258, in cuda
   return self._apply(lambda t: t.cuda(device))
   File "/home/dl/anaconda2/envs/test_py3/lib/python3.6/site-packages/torch/nn/modules/module.py", line 185, in _apply
   module._apply(fn)
   File "/home/dl/anaconda2/envs/test_py3/lib/python3.6/site-packages/torch/nn/modules/module.py", line 185, in _apply
   module._apply(fn)
   File "/home/dl/anaconda2/envs/test_py3/lib/python3.6/site-packages/torch/nn/modules/module.py", line 185, in _apply
   module._apply(fn)
   File "/home/dl/anaconda2/envs/test_py3/lib/python3.6/site-packages/torch/nn/modules/module.py", line 191, in _apply
   param.data = fn(param.data)
   File "/home/dl/anaconda2/envs/test_py3/lib/python3.6/site-packages/torch/nn/modules/module.py", line 258, in
   return self._apply(lambda t: t.cuda(device))

i don't know what to do ...

Thanks for your impressive job，can you tell me what is the name of software you drawed network ？.

I don't know what that means. Can you elaborate on the specific function of this piece? Why use the 3x3 convolution kernel?

self.score = nn.Sequential(OrderedDict([("norm.1", norm_act(out_sec + in_stag2_up_chs)),
("conv.1", nn.Conv2d(out_sec + in_stag2_up_chs,
out_sec + in_stag2_up_chs,
kernel_size=3, stride=1, padding=2,
dilation=2, bias=False)),
("norm.2", norm_act(out_sec + in_stag2_up_chs)),
("conv.2", nn.Conv2d(out_sec + in_stag2_up_chs, self.n_class,
kernel_size=1, stride=1, padding=0,
bias=True)),
("up1", nn.Upsample(size=in_size, mode='bilinear'))]))

Hi,
Thank you for your great work! I am trying to do some distillation work on the small mobile net. I have some questions:
The code need the package pytorch encoding, but the pytorch encoding requires the pytorch to be 0.5.0. Your code is about pytorch 0.3.0+. I wonder how do you install the package of pytorch encoding?

However, the code can run under pytorch 0.5.0, but the training speed is very slow. I used a single P100 to have a try, and it need nearly 7 minutes to finish one epoch. If we need to train 10000 epochs, it will take nearly 48 days.
Is this because the different pytorch version?

Thanks,
Yifan

The following error occurred when I ran the train_mobile_mvd:
/workfs/lhhuang/anaconda3/envs/pytorch0.3/lib/python3.6/site-packages/torch/cuda/init.py:95: UserWarning:
Found GPU0 Tesla V100-SXM2-16GB which requires CUDA_VERSION >= 9000 for
optimal performance and fast startup time, but your PyTorch was compiled
with CUDA_VERSION 8000. Please install the correct PyTorch binary
using instructions from http://pytorch.org

warnings.warn(incorrect_binary_warn % (d, name, 9000, CUDA_VERSION))
^M 0%| | 0/9000 [00:00<?, ?it/s]['/workfs/lhhuang/segment/LightNet-master2/LightNet-master/scripts', '/workfs/lhhuang/segment/LightNet-master2/LightNet-master/scripts', '/dat01/lhhuang/models/research', '/dat01/lhhuang/models/research/slim', '/dat01/lhhuang/models/research/delf', '/workfs/lhhuang/anaconda3/envs/pytorch0.3/lib/python36.zip', '/workfs/lhhuang/anaconda3/envs/pytorch0.3/lib/python3.6', '/workfs/lhhuang/anaconda3/envs/pytorch0.3/lib/python3.6/lib-dynload', '/home/lhhuang/.local/lib/python3.6/site-packages', '/workfs/lhhuang/anaconda3/envs/pytorch0.3/lib/python3.6/site-packages', '/workfs/lhhuang/segment/LightNet-master2/LightNet-master/datasets', '/workfs/lhhuang/segment/LightNet-master2/LightNet-master']

+++++++++++++++++++++++++++++++++++++++++++++++++++++++

0. Setting up DataLoader...

+++++++++++++++++++++++++++++++++++++++++++++++++++++++

There are 66 labels in the config file
Found 18000 training images...
There are 66 labels in the config file
Found 2000 validation images...

+++++++++++++++++++++++++++++++++++++++++++++++++++++++

1. Setting up Model...
2. Model Training start...
   ^M 0%| | 1/9000 [00:04<12:23:13, 4.96s/it]^M> Epoch [1/256]: 0%| | 1/9000 [00:04<12:23:13, 4.96s/it]Traceback (most recent call last):
   File "train_mobile_mvd.py", line 348, in
   train(train_args, data_path, save_path)
   File "train_mobile_mvd.py", line 185, in train
   out_stg1, out_stg2, out_stg3 = model(images) # Here we have 3 output for 3 loss
   File "/workfs/lhhuang/anaconda3/envs/pytorch0.3/lib/python3.6/site-packages/torch/nn/modules/module.py", line 357, in call
   result = self.forward(*input, **kwargs)
   File "/workfs/lhhuang/segment/LightNet-master2/LightNet-master/models/mobilenetv2plus.py", line 169, in forward
   de_stg1 = self.aspp(stg8)[1]
   File "/workfs/lhhuang/anaconda3/envs/pytorch0.3/lib/python3.6/site-packages/torch/nn/modules/module.py", line 357, in call
   result = self.forward(*input, **kwargs)
   File "/workfs/lhhuang/anaconda3/envs/pytorch0.3/lib/python3.6/site-packages/torch/nn/modules/container.py", line 67, in forward
   input = module(input)
   File "/workfs/lhhuang/anaconda3/envs/pytorch0.3/lib/python3.6/site-packages/torch/nn/modules/module.py", line 357, in call
   result = self.forward(*input, **kwargs)
   File "/workfs/lhhuang/segment/LightNet-master2/LightNet-master/modules/misc.py", line 334, in forward
   x = self.in_norm(x)
   File "/workfs/lhhuang/anaconda3/envs/pytorch0.3/lib/python3.6/site-packages/torch/nn/modules/module.py", line 357, in call
   result = self.forward(*input, **kwargs)
   File "/workfs/lhhuang/segment/LightNet-master2/LightNet-master/modules/bn.py", line 96, in forward
   self.activation, self.slope)
   File "/workfs/lhhuang/segment/LightNet-master2/LightNet-master/modules/functions.py", line 97, in forward
   _act_forward(ctx, x)
   File "/workfs/lhhuang/segment/LightNet-master2/LightNet-master/modules/functions.py", line 41, in _act_forward
   _ext.leaky_relu_cuda(x, ctx.slope)
   File "/workfs/lhhuang/anaconda3/envs/pytorch0.3/lib/python3.6/site-packages/torch/utils/ffi/init.py", line 180, in safe_call
   result = torch._C._safe_call(*args, **kwargs)
   torch.FatalError: function_attributes(): after cudaFuncGetAttributes: invalid device function

May I ask how to train model with both gtFine and gtCoarse dataset? And another question, how to get the model mIOU when test on val dataset?Thank you !

I have build pytorch from source with CUDA 8 and cudnn 7 and installed pytroch-Encoding succesfully.
But when I try to train the model, it shows the following error.

➜ LightNet git:(master) ✗ python scripts/train_mobile.py

+++++++++++++++++++++++++++++++++++++++++++++++++++++++

0. Setting up DataLoader...

+++++++++++++++++++++++++++++++++++++++++++++++++++++++

Found 2975 train images...
Found 500 val images...

+++++++++++++++++++++++++++++++++++++++++++++++++++++++

1. Setting up Model...
   THCudaCheck FAIL file=/home/kenzhang/pytorch/aten/src/THC/THCGeneral.cpp line=70 error=38 : no CUDA-capable device is detected
   Traceback (most recent call last):
   File "scripts/train_mobile.py", line 372, in
   train(train_args, data_path, save_path)
   File "scripts/train_mobile.py", line 75, in train
   model = torch.nn.DataParallel(model, device_ids=[0, 1]).cuda()
   File "/opt/miniconda3/lib/python3.6/site-packages/torch/nn/modules/module.py", line 230, in cuda
   return self._apply(lambda t: t.cuda(device))
   File "/opt/miniconda3/lib/python3.6/site-packages/torch/nn/modules/module.py", line 152, in _apply
   module._apply(fn)
   File "/opt/miniconda3/lib/python3.6/site-packages/torch/nn/modules/module.py", line 152, in _apply
   module._apply(fn)
   File "/opt/miniconda3/lib/python3.6/site-packages/torch/nn/modules/module.py", line 152, in _apply
   module._apply(fn)
   File "/opt/miniconda3/lib/python3.6/site-packages/torch/nn/modules/module.py", line 158, in _apply
   param.data = fn(param.data)
   File "/opt/miniconda3/lib/python3.6/site-packages/torch/nn/modules/module.py", line 230, in
   return self._apply(lambda t: t.cuda(device))
   RuntimeError: cuda runtime error (38) : no CUDA-capable device is detected at /home/kenzhang/pytorch/aten/src/THC/THCGeneral.cpp:70

in LightNet-master\models\mobilenetv2exfuse.py

line 184: if self.traval == "train" and self.n_class != 0:
line 208: elif self.traval == "train" and self.n_class != 0:

why a repetition?
Thanks for reply!

using pytorch1.0 as required by pytorch.encoding

sofa, nice job

Hello Authors,

I am encountering this issue, and am not able to solve this.

Traceback (most recent call last): File "scripts/train_mobile_scannet.py", line 22, in <module> from models.rfmobilenetv2plus import RFMobileNetV2Plus File "/nfs/home/us000146/LightNet/models/rfmobilenetv2plus.py", line 6, in <module> from modules import SCSEBlock, InPlaceABN, InPlaceABNWrapper, RFBlock File "/nfs/home/us000146/LightNet/modules/__init__.py", line 4, in <module> from . context_encode import ContextEncodeInplaceABN, ContextEncodeDropInplaceABN File "/nfs/home/us000146/LightNet/modules/context_encode.py", line 4, in <module> import encoding ModuleNotFoundError: No module named 'encoding'

I am running under the following configuration:

Python 3.6.13
GCC 7.3.0 [on linux]
Torch 0.4.1.post2

I have tried a few things but nothing works out and I am stuck here, without being able to train the model. Please suggest what should I do?

Regards,
Nitin Bansal

I want to use the vistas dataset for semantic segmentation training, but the official provide labelids images with only instance segmentation, and no labelids images with semantic segmentation. What should I do?

I can only get miou below 0.4 when training from scratch without GAN augmentation

Hello, I wonder if I can use my own dataset to train on these networks? And is there any limitation about the objects' classes? For example, some housewares. If the segmentation task on them can be done, could you please tell me the pipeline? I have no idea how to organize my data into a proper form to fit in the code. Thanks a lot!

hi,I have trained SE-WResNetV2 about 100 epochs, i get mIOU=0.756, it is far away from 0.8013 mIOU which you've get and share, en , i want to known there is some tricks about train SE-WResNetV2? And how to get the same mIOU with you?

Excuse me but, I'd like to ask about the RFMobilenetV2Plus pretrained model. When will you upload those pretrained model (either pretrained on imagenet or cityscape)?
I didnt's find any clue in your README.md
Thanks.