hi, when i using resume to load parameters, it is crash saying ''unexpected key "module.features.0.weight" in state_dict''. I don't know what wrong with this, Thanks a lot!

Hi,

I want to test on the example cases, so I just run python example/voc/train_fcn8s.py -g 1 on a DGX station, with Tesla V100 16GB memory. After loading the model, it immediately reports out of memory in the first forward pass. Has anyone encountered this situation?

environment: Pytorch: v0.4.1 CUDA: 9.0 cuDNN: 7.1.4

Kaicheng

Hello, I am new to this community and I have doubt regarding the fcn32.py. It's getting hard for me to understand that the output of self.upscore is [batch_size,21,64,64], then how it doesn't generate any error while calculating the loss with the ground-truth. According to me, the output should be [batch_size,21,64,64] and then it should be rescaled to size [batch_size,21,224,224] If your input image is of 224x224x3 size. Am I right over here ? How is this upsampling taking place ? how does our self.upscore output size matches with ground-truth size ?

The image is correctly segmented when I run FCN in caffe. I recently decided to switch to pytorch, and installed pytorch FCN. Here's the self-contained code:

import torch
import sys, os, path
# this sets the same order as in nvidia-smi
os.environ['CUDA_DEVICE_ORDER'] = 'PCI_BUS_ID'
os.environ['CUDA_VISIBLE_DEVICES'] = '1'
import numpy as np
import skimage.data
from PIL import Image
import matplotlib.pyplot as plt
import torch
from torch import unsqueeze
from torch.autograd import Variable
from torch.utils.data import Dataset, DataLoader, TensorDataset
from torch.utils.data.dataset import Subset
from torchvision.datasets import ImageFolder
from torchvision.transforms import Compose, RandomResizedCrop, RandomVerticalFlip, RandomHorizontalFlip
from torchvision.transforms import ColorJitter, ToTensor, Normalize, ToPILImage, Resize
from torchvision.utils import save_image
# torch for FCN
import torchfcn
##########################################
#
# Compose: combine a number of image transforms
#
########################################## 


img = "000000559099.png"
device = "cuda"

def preprocess (img):
    im = Image.open(img)
    norm = Normalize(mean = [0.485, 0.456, 0.406], std = [0.229, 0.224, 0.225])
    transf = Compose ([Resize((300,300)), ToTensor(), norm])
    im = transf(im)  
    im = im.unsqueeze(0) 
    return im

im = preprocess(img)

print 'img', im.shape
print torch.cuda.get_device_name(0)
print torch.cuda.get_device_capability(0)
# set to Tesla K40m
torch.cuda.device(1)
print 'CURRENT',  torch.cuda.current_device()
model = torchfcn.models.FCN8s(n_class=21)
model = model.cuda()
im = im.cuda()
im = Variable(im, requires_grad = False)
weights = torch.load('/home/ICTDOMAIN/453615/data/models/pytorch/fcn8s_from_caffe.pth')
model.load_state_dict(weights)
print (model)
score = model(im)	
output= score.data.cpu().numpy()
print output.shape
plt.imshow(output[0].argmax(0))
plt.show()

Every pixel is classified as background. What is wrong with the code?

when i run the command: pip install torchfcn i got the error like this. I have tried re-install torch, but the bug is still here.

Log: Installing collected packages: torchfcn Running setup.py install for torchfcn ... error Complete output from command /usr/bin/python -u -c "import setuptools, tokenize;file='/tmp/pip-install-EVS88p/torchfcn/setup.py';f=getattr(tokenize, 'open', open)(file);code=f.read().replace('\r\n', '\n');f.close();exec(compile(code, file, 'exec'))" install --record /tmp/pip-record-W87s66/install-record.txt --single-version-externally-managed --compile: running install

 error: can't copy 'torchfcn/ext/fcn.berkeleyvision.org': doesn't exist or not a regular file
    
    ----------------------------------------
Command "/usr/bin/python -u -c "import setuptools, tokenize;__file__='/tmp/pip-install-EVS88p/torchfcn/setup.py';f=getattr(tokenize, 'open', open)(__file__);code=f.read().replace('\r\n', '\n');f.close();exec(compile(code, __file__, 'exec'))" install --record /tmp/pip-record-W87s66/install-record.txt --single-version-externally-managed --compile" failed with error code 1 in /tmp/pip-install-EVS88p/torchfcn/
chao@user-Server:~/chao/action/segmentation/pytorch-fcn$ sudo pip uninstall torchfcn

Hi, thank you for sharing the code. I trained FCN32s on VOC12 and got the correct results when using view_log. However, when I tried to draw the learning curves, my log couldn't work with the code. After rolling_mean , all the data in df_train became NaN. Is there something wrong with my log? There are some pics about this problem. Do you know how to fix it? Thank you very much. @wkentaro

When I test the trained model use the test data or the train data ,the output is only zero.Anyone can help me to solve this?Thanks! the output like this: [[0 0 0 ..., 0 0 0] [0 0 0 ..., 0 0 0] [0 0 0 ..., 0 0 0] ..., [0 0 0 ..., 0 0 0] [0 0 0 ..., 0 0 0] [0 0 0 ..., 0 0 0]]

When I follow the voc examples, I got problems of visualizing the logs in command ./view_log logs/XXX/log.csv with error

can't multiply sequence by non-int of type 'float'

Hello, wkentaro. Do you find that the training loss will suddenly become larger when a new epoch starts? I've noticed that and think it is because of the incorrect use of the training and evaluating mode of the model. In train_epoch(), line 171, when you finish validation, I think you should change the model back into training mode.

Hi @wkentaro , I see that you use the vgg16 pretrained model from caffe. I wonder that why you do not use a pretrained one provided by torchvision. Is the pretrained one from caffe better?

Thanks.

When I implement the VOC2012 Dataset based your code, it occurs an error "incosistent tensor size". The code about dataset is same with yours. But I add some test code as below:

if __name__ == '__main__':
    dst = VOCDataSet("./data", is_transform=True)
    trainloader = data.DataLoader(dst, batch_size=4)
    for i, data in enumerate(trainloader):
        imgs, labels = data
        print(imgs[0].type())

Did you encounter this problem?

I wanted to change the initialization method of the convolution from 0 to a positive-terrestrial distribution, but this led to a gradient explosion after a few iters.Why does this happen?

Traceback (most recent call last): File "demo.py", line 59, in demo(args) File "demo.py", line 45, in demo model = get_model(args.model, local_rank=args.local_rank, pretrained=True, root=args.save_folder).to(device) File "E:\Python_Projects\awesome-semantic-segmentation-pytorch-master\core\models\model_zoo.py", line 83, in get_model net = _modelsname File "E:\Python_Projects\awesome-semantic-segmentation-pytorch-master\core\models\fcn.py", line 219, in get_fcn32s_vgg16_voc return get_fcn32s('pascal_voc', 'vgg16', **kwargs) File "E:\Python_Projects\awesome-semantic-segmentation-pytorch-master\core\models\fcn.py", line 172, in get_fcn32s model.load_state_dict(torch.load("E:\fcn32s_vgg16_pascal_voc.pth",map_location="cpu")) File "D:\Anaconda3\lib\site-packages\torch\nn\modules\module.py", line 1483, in load_state_dict self.class.name, "\n\t".join(error_msgs))) RuntimeError: Error(s) in loading state_dict for FCN32s: Missing key(s) in state_dict: "pretrained.0.weight", "pretrained.0.bias", "pretrained.2.weight", "pretrained.2.bias", "pretrained.5.weight", "pretrained.5.bias", "pretrained.7.weight", "pretrained.7.bias", "pretrained.10.weight", "pretrained.10.bias", "pretrained.12.weight", "pretrained.12.bias", "pretrained.14.weight", "pretrained.14.bias", "pretrained.17.weight", "pretrained.17.bias", "pretrained.19.weight", "pretrained.19.bias", "pretrained.21.weight", "pretrained.21.bias", "pretrained.24.weight", "pretrained.24.bias", "pretrained.26.weight", "pretrained.26.bias", "pretrained.28.weight", "pretrained.28.bias", "head.block.0.weight", "head.block.1.weight", "head.block.1.bias", "head.block.1.running_mean", "head.block.1.running_var", "head.block.4.weight", "head.block.4.bias". Unexpected key(s) in state_dict: "features.0.weight", "features.0.bias", "features.2.weight", "features.2.bias", "features.5.weight", "features.5.bias", "features.7.weight", "features.7.bias", "features.10.weight", "features.10.bias", "features.12.weight", "features.12.bias", "features.14.weight", "features.14.bias", "features.17.weight", "features.17.bias", "features.19.weight", "features.19.bias", "features.21.weight", "features.21.bias", "features.24.weight", "features.24.bias", "features.26.weight", "features.26.bias", "features.28.weight", "features.28.bias", "classifier.0.weight", "classifier.0.bias", "classifier.3.weight", "classifier.3.bias", "classifier.6.weight", "classifier.6.bias".

I wonder why the padding of first conv layer in FCN32s is 100? FCN32s( (conv1_1): Conv2d(3, 64, kernel_size=(3, 3), stride=(1, 1), padding=(100, 100)) I am a newcomer in ML.Can you give me some answer? Thank you for advance

When I ran ./train_fcn32s.py -g 0，I got this error Traceback (most recent call last): File "/home/jbwen/anaconda3/envs/py36/lib/python3.6/site-packages/urllib3/connection.py", line 160, in _new_conn (self._dns_host, self.port), self.timeout, **extra_kw File "/home/jbwen/anaconda3/envs/py36/lib/python3.6/site-packages/urllib3/util/connection.py", line 84, in create_connection raise err File "/home/jbwen/anaconda3/envs/py36/lib/python3.6/site-packages/urllib3/util/connection.py", line 74, in create_connection sock.connect(sa) OSError: [Errno 101] Network is unreachable

During handling of the above exception, another exception occurred:

Traceback (most recent call last): File "/home/jbwen/anaconda3/envs/py36/lib/python3.6/site-packages/urllib3/connectionpool.py", line 677, in urlopen chunked=chunked, File "/home/jbwen/anaconda3/envs/py36/lib/python3.6/site-packages/urllib3/connectionpool.py", line 392, in _make_request conn.request(method, url, **httplib_request_kw) File "/home/jbwen/anaconda3/envs/py36/lib/python3.6/http/client.py", line 1262, in request self._send_request(method, url, body, headers, encode_chunked) File "/home/jbwen/anaconda3/envs/py36/lib/python3.6/http/client.py", line 1308, in _send_request self.endheaders(body, encode_chunked=encode_chunked) File "/home/jbwen/anaconda3/envs/py36/lib/python3.6/http/client.py", line 1257, in endheaders self._send_output(message_body, encode_chunked=encode_chunked) File "/home/jbwen/anaconda3/envs/py36/lib/python3.6/http/client.py", line 1036, in _send_output self.send(msg) File "/home/jbwen/anaconda3/envs/py36/lib/python3.6/http/client.py", line 974, in send self.connect() File "/home/jbwen/anaconda3/envs/py36/lib/python3.6/site-packages/urllib3/connection.py", line 187, in connect conn = self._new_conn() File "/home/jbwen/anaconda3/envs/py36/lib/python3.6/site-packages/urllib3/connection.py", line 172, in _new_conn self, "Failed to establish a new connection: %s" % e urllib3.exceptions.NewConnectionError: <urllib3.connection.HTTPConnection object at 0x7f1ef9d08588>: Failed to establish a new connection: [Errno 101] Network is unreachable

During handling of the above exception, another exception occurred:

Traceback (most recent call last): File "/home/jbwen/anaconda3/envs/py36/lib/python3.6/site-packages/requests/adapters.py", line 449, in send timeout=timeout File "/home/jbwen/anaconda3/envs/py36/lib/python3.6/site-packages/urllib3/connectionpool.py", line 725, in urlopen method, url, error=e, _pool=self, _stacktrace=sys.exc_info()[2] File "/home/jbwen/anaconda3/envs/py36/lib/python3.6/site-packages/urllib3/util/retry.py", line 439, in increment raise MaxRetryError(_pool, url, error or ResponseError(cause)) urllib3.exceptions.MaxRetryError: HTTPConnectionPool(host='drive.google.com', port=80): Max retries exceeded with url: /uc?id=0B9P1L--7Wd2vLTJZMXpIRkVVRFk (Caused by NewConnectionError('<urllib3.connection.HTTPConnection object at 0x7f1ef9d08588>: Failed to establish a new connection: [Errno 101] Network is unreachable',))

During handling of the above exception, another exception occurred:

Traceback (most recent call last): File "./train_fcn32s.py", line 151, in main() File "./train_fcn32s.py", line 116, in main vgg16 = torchfcn.models.VGG16(pretrained=True) File "/home/jbwen/anaconda3/envs/py36/lib/python3.6/site-packages/torchfcn/models/vgg.py", line 13, in VGG16 model_file = _get_vgg16_pretrained_model() File "/home/jbwen/anaconda3/envs/py36/lib/python3.6/site-packages/torchfcn/models/vgg.py", line 23, in _get_vgg16_pretrained_model md5='aa75b158f4181e7f6230029eb96c1b13', File "/home/jbwen/anaconda3/envs/py36/lib/python3.6/site-packages/fcn/data.py", line 32, in cached_download gdown.download(url, path, quiet=quiet) File "/home/jbwen/anaconda3/envs/py36/lib/python3.6/site-packages/gdown/download.py", line 110, in download res = sess.get(url, stream=True) File "/home/jbwen/anaconda3/envs/py36/lib/python3.6/site-packages/requests/sessions.py", line 543, in get return self.request('GET', url, **kwargs) File "/home/jbwen/anaconda3/envs/py36/lib/python3.6/site-packages/requests/sessions.py", line 530, in request resp = self.send(prep, **send_kwargs) File "/home/jbwen/anaconda3/envs/py36/lib/python3.6/site-packages/requests/sessions.py", line 643, in send r = adapter.send(request, **kwargs) File "/home/jbwen/anaconda3/envs/py36/lib/python3.6/site-packages/requests/adapters.py", line 516, in send raise ConnectionError(e, request=request) requests.exceptions.ConnectionError: HTTPConnectionPool(host='drive.google.com', port=80): Max retries exceeded with url: /uc?id=0B9P1L--7Wd2vLTJZMXpIRkVVRFk (Caused by NewConnectionError('<urllib3.connection.HTTPConnection object at 0x7f1ef9d08588>: Failed to establish a new connection: [Errno 101] Network is unreachable',))