Hi, there
You provided roughly trained weight in this repo, but the performance was not so good.
To get the accuracy mentioned in your paper, can you provide the well trained model of RoadNet?

Thanks

Hello, I am using my own set of images to train the RoadNet model. I marked it as follows.

During training, the target image of visdom only shows a black colour. So can you please tell me, what are those mistakes I might have made?

Hi! I find that the crack segmentation results downloaded from google drive are much better than those in your paper. For example, iou is improved from 85.9 to 87.8. What modifications did you make after paper publication?

hi
How do I make dataset labels?

Hello there,

I forked the project to create a web app deployed on Heroku free tier to be easy to use as a DEMO of your work.

Some considerations:

• Only cpu inference (image[png/jpg/jpeg/bmp] and video[mp4] allowed)
• Automatic resizing input images and video frames (500px).
• Only DeepCrack
• "_fused" images taken as output

Web version https://deepcrackcpu.herokuapp.com/
Repo of web version https://github.com/DZDL/crack-detector

Hi @yhlleo

When I used eval.py and I got the output file, how I know which bT is the best?

Could you guide me?

Many thanks

Hi，where is the codes about he ODS,OIS and AP?

Thanks a ton for your work – I try to create a path from the detected edges to compare it with Open Streetmap Data. However I cannot find out where the edges are created or how to work with them. Can I get the edge pixels separated from the "road surface" and the rest? Same applies to centerline which I'd like to separate from the rest of the result.

I am aware that I will not get a path but I want a result similar to

Thank you for your code. I met this error when I retrain deepcrack model : output with device cuda:0 and dtype Long doesn't match the desired device cuda:0 and dtype Float. Could you give me some advises? thanks.

Hello,

First of all thank you very much for releasing the code for the community!

I am facing some problems when trying to test the DeepCrack model. I have downloaded the deepcrack dataset and installed the environment correctly. However, when testing the model the predicted images (side and fused images) are in light gray color. Despite, when loading the model weights provided, I am facing the following issue:

dataset [DeepCrackDataset] was created
initialize network with xavier
model [DeepCrackModel] was created
loading the model from ./checkpoints/deepcrack/pretrained_net_G.pth
Ignoring "Error(s) in loading state_dict for DeepCrackNet:
Unexpected key(s) in state_dict: "conv1.1.weight", "conv1.1.bias", "conv1.4.weight", "conv1.4.bias", "conv2.1.weight", "conv2.1.bias", "conv2.4.weight", "conv2.4.bias", "conv3.1.weight", "conv3.1.bias", "conv3.4.weight", "conv3.4.bias", "conv3.7.weight", "conv3.7.bias", "conv4.1.weight", "conv4.1.bias", "conv4.4.weight", "conv4.4.bias", "conv4.7.weight", "conv4.7.bias", "conv5.1.weight", "conv5.1.bias", "conv5.4.weight", "conv5.4.bias", "conv5.7.weight", "conv5.7.bias".
size mismatch for side_conv1.weight: copying a param of torch.Size([2, 64, 1, 1]) from checkpoint, where the shape is torch.Size([1, 64, 1, 1]) in current model.
size mismatch for side_conv2.weight: copying a param of torch.Size([2, 128, 1, 1]) from checkpoint, where the shape is torch.Size([1, 128, 1, 1]) in current model.
size mismatch for side_conv3.weight: copying a param of torch.Size([2, 256, 1, 1]) from checkpoint, where the shape is torch.Size([1, 256, 1, 1]) in current model.
size mismatch for side_conv4.weight: copying a param of torch.Size([2, 512, 1, 1]) from checkpoint, where the shape is torch.Size([1, 512, 1, 1]) in current model.
size mismatch for side_conv5.weight: copying a param of torch.Size([2, 512, 1, 1]) from checkpoint, where the shape is torch.Size([1, 512, 1, 1]) in current model.
size mismatch for fuse_conv.weight: copying a param of torch.Size([2, 10, 1, 1]) from checkpoint, where the shape is torch.Size([1, 5, 1, 1]) in current model."
---------- Networks initialized -------------
[Network G] Total number of parameters : 14.713 M

Could you please help me?

After running the test on the crack dataset the processed images in the ./results/deepcrack/test_latest/images folder are incorrect (or shifted or locked).

Hello there I created a simple Colab Notebook to run DeepCrack inference

https://colab.research.google.com/drive/1t2C5mBOy7xuB4kP9TQzVOI0nglJBkMMD?usp=sharing

Please consider to add to your docs, thanks in advance

我想问一个很简单的问题，但这对我来说很困难，我还没有成功运行这个程序
我做了许多尝试比如
visdom的端口号，deault设置为8123（一开始我使用的默认）
自己创建了一个results文件夹，
但是使用sh ./scripts/train_roadnet.sh 0后，虽然GPU可以工作
但是不知道训练的结果在哪里，在checkpoint里面也没有新的数据
只会在DeepSegmentor/checkpoints/roadnet/loss_log多一行如下的数据
================ Training Loss (Tue Apr 12 20:43:56 2022) ================
请问还需要怎样做呢

Hi,
@yhlleo ,
Thanks a lot for your work!
I would like to know if you omit the evaluation process during training intentionally?
To my knowledge, The evaluation (differs from test procedure) process is used to evaluate the performance of the network during training, and then select the model weights that got the best performance. So is there any reason to omit the process?

ValueError: Target size (torch.Size([1, 1, 384, 544])) must be the same as input size (torch.Size([1, 2, 384, 544]))

File: base_model.py Line 197

Error in net.load_state_dict(state_dict)

dataset [RoadNetDataset] was created
initialize network with xavier
model [RoadNetModel] was created
loading the model from ./checkpoints/roadnet/pretrained_net_G.pth
Traceback (most recent call last):
File "test.py", line 20, in
model.setup(opt) # regular setup: load and print networks; create schedulers
File "/home/ibrahim/Desktop/Research/DeepSegmentor/models/base_model.py", line 88, in setup
self.load_networks(load_suffix)
File "/home/ibrahim/Desktop/Research/DeepSegmentor/models/base_model.py", line 197, in load_networks
net.load_state_dict(state_dict)
File "/home/ibrahim/anaconda3/envs/env-yolo/lib/python3.7/site-packages/torch/nn/modules/module.py", line 847, in load_state_dict
self.class.name, "\n\t".join(error_msgs)))
RuntimeError: Error(s) in loading state_dict for RoadNet:
Unexpected key(s) in state_dict: "segment_conv1.1.weight", "segment_conv1.1.bias", "segment_conv1.4.weight", "segment_conv1.4.bias", "segment_conv2.1.weight", "segment_conv2.1.bias", "segment_conv2.4.weight", "segment_conv2.4.bias", "segment_conv3.1.weight", "segment_conv3.1.bias", "segment_conv3.4.weight", "segment_conv3.4.bias", "segment_conv3.7.weight", "segment_conv3.7.bias", "segment_conv4.1.weight", "segment_conv4.1.bias", "segment_conv4.4.weight", "segment_conv4.4.bias", "segment_conv4.7.weight", "segment_conv4.7.bias", "segment_conv5.1.weight", "segment_conv5.1.bias", "segment_conv5.4.weight", "segment_conv5.4.bias", "segment_conv5.7.weight", "segment_conv5.7.bias", "edge_conv1.1.weight", "edge_conv1.1.bias", "edge_conv1.4.weight", "edge_conv1.4.bias", "edge_conv2.1.weight", "edge_conv2.1.bias", "edge_conv2.4.weight", "edge_conv2.4.bias", "edge_conv3.1.weight", "edge_conv3.1.bias", "edge_conv3.4.weight", "edge_conv3.4.bias", "edge_conv4.1.weight", "edge_conv4.1.bias", "edge_conv4.4.weight", "edge_conv4.4.bias", "centerline_conv1.1.weight", "centerline_conv1.1.bias", "centerline_conv1.4.weight", "centerline_conv1.4.bias", "centerline_conv2.1.weight", "centerline_conv2.1.bias", "centerline_conv2.4.weight", "centerline_conv2.4.bias", "centerline_conv3.1.weight", "centerline_conv3.1.bias", "centerline_conv3.4.weight", "centerline_conv3.4.bias", "centerline_conv4.1.weight", "centerline_conv4.1.bias", "centerline_conv4.4.weight", "centerline_conv4.4.bias".
size mismatch for side_segment_conv1.weight: copying a param with shape torch.Size([1, 64, 1, 1]) from checkpoint, the shape in current model is torch.Size([3, 64, 1, 1]).
size mismatch for side_segment_conv2.weight: copying a param with shape torch.Size([1, 128, 1, 1]) from checkpoint, the shape in current model is torch.Size([3, 128, 1, 1]).
size mismatch for side_segment_conv3.weight: copying a param with shape torch.Size([1, 256, 1, 1]) from checkpoint, the shape in current model is torch.Size([3, 256, 1, 1]).
size mismatch for side_segment_conv4.weight: copying a param with shape torch.Size([1, 512, 1, 1]) from checkpoint, the shape in current model is torch.Size([3, 512, 1, 1]).
size mismatch for side_segment_conv5.weight: copying a param with shape torch.Size([1, 512, 1, 1]) from checkpoint, the shape in current model is torch.Size([3, 512, 1, 1]).
size mismatch for fuse_segment_conv.weight: copying a param with shape torch.Size([1, 5, 1, 1]) from checkpoint, the shape in current model is torch.Size([3, 15, 1, 1]).
size mismatch for edge_conv1.0.weight: copying a param with shape torch.Size([32, 4, 3, 3]) from checkpoint, the shape in current model is torch.Size([32, 6, 3, 3]).
size mismatch for side_edge_conv1.weight: copying a param with shape torch.Size([1, 32, 1, 1]) from checkpoint, the shape in current model is torch.Size([3, 32, 1, 1]).
size mismatch for side_edge_conv2.weight: copying a param with shape torch.Size([1, 64, 1, 1]) from checkpoint, the shape in current model is torch.Size([3, 64, 1, 1]).
size mismatch for side_edge_conv3.weight: copying a param with shape torch.Size([1, 128, 1, 1]) from checkpoint, the shape in current model is torch.Size([3, 128, 1, 1]).
size mismatch for side_edge_conv4.weight: copying a param with shape torch.Size([1, 256, 1, 1]) from checkpoint, the shape in current model is torch.Size([3, 256, 1, 1]).
size mismatch for fuse_edge_conv.weight: copying a param with shape torch.Size([1, 4, 1, 1]) from checkpoint, the shape in current model is torch.Size([3, 12, 1, 1]).
size mismatch for centerline_conv1.0.weight: copying a param with shape torch.Size([32, 4, 3, 3]) from checkpoint, the shape in current model is torch.Size([32, 6, 3, 3]).
size mismatch for side_centerline_conv1.weight: copying a param with shape torch.Size([1, 32, 1, 1]) from checkpoint, the shape in current model is torch.Size([3, 32, 1, 1]).
size mismatch for side_centerline_conv2.weight: copying a param with shape torch.Size([1, 64, 1, 1]) from checkpoint, the shape in current model is torch.Size([3, 64, 1, 1]).
size mismatch for side_centerline_conv3.weight: copying a param with shape torch.Size([1, 128, 1, 1]) from checkpoint, the shape in current model is torch.Size([3, 128, 1, 1]).
size mismatch for side_centerline_conv4.weight: copying a param with shape torch.Size([1, 256, 1, 1]) from checkpoint, the shape in current model is torch.Size([3, 256, 1, 1]).
size mismatch for fuse_centerline_conv.weight: copying a param with shape torch.Size([1, 4, 1, 1]) from checkpoint, the shape in current model is torch.Size([3, 12, 1, 1]).

Hi,

Thanks a lot for providing this very useful code. I have a question about your code. Because I saw you used different thresholds to binarize the predicted image, I wonder how the prediction is changed to an image? prediction of the model -> sigmoid -> *255 to array -> image? Thanks!

Hi there,
I'm trying to build my own CrackDetector. In the step of Weights balancing, what is path_to_segmentation ?

Sorry for the eventually dumb question.

roadnet评估时，用的这个指令：python eval.py --metric_mode sem --model_name latest_net_G.pth --output roadnet.prf，出现了报错。请问这个指令是这样修改的么

Hi,
Good work. This code is for binary segmentation. If we have 4 or 5 classes in dataset then how can we calculate the weights of each class?

I'm training on my own dataset the DeepCrack segmentor.

But when I calcuate the loss function for bp with model.optimize_parameters(epoch) # calculate loss functions, get gradients, update network weights, it raises me an error on shapes:

ValueError: Target size (torch.Size([1, 1, 155, 420])) must be the same as input size (torch.Size([1, 2, 155, 420]))

how to solve this?

Error(s) in loading state_dict for RoadNet
For roadnet I have download the pretrained model form https://pan.baidu.com/s/1GEevUwp1xrEtASjixawrwg
parameters of the model defied in the code:

parameters in the pretrained model:

the model in the code seems missing some 1*1 conv layers.
May you have a look at this? thanks!

Traceback (most recent call last):
  File "train.py", line 32, in <module>
    model.optimize_parameters(epoch)   # calculate loss functions, get gradients, update network weights
  File "/rds/project/t2_vol2/rds-t2-cs056/ll610/Onepiece/zh_data/DeepSegmentor/models/deepcrack_model.py", line 115, in optimize_parameters
    self.backward()             # calculate gradients for G
  File "/rds/project/t2_vol2/rds-t2-cs056/ll610/Onepiece/zh_data/DeepSegmentor/models/deepcrack_model.py", line 103, in backward
    self.loss_side += self.criterionSeg(out, self.label) * w
  File "/home/ll610/miniconda3/envs/deepsegmentor/lib/python3.5/site-packages/torch/nn/modules/module.py", line 477, in __call__
    result = self.forward(*input, **kwargs)
  File "/rds/project/t2_vol2/rds-t2-cs056/ll610/Onepiece/zh_data/DeepSegmentor/models/deepcrack_networks.py", line 103, in forward
    BCE_loss = self.criterion(inputs, targets)
  File "/home/ll610/miniconda3/envs/deepsegmentor/lib/python3.5/site-packages/torch/nn/modules/module.py", line 477, in __call__
    result = self.forward(*input, **kwargs)
  File "/home/ll610/miniconda3/envs/deepsegmentor/lib/python3.5/site-packages/torch/nn/modules/loss.py", line 573, in forward
    reduction=self.reduction)
  File "/home/ll610/miniconda3/envs/deepsegmentor/lib/python3.5/site-packages/torch/nn/functional.py", line 1651, in binary_cross_entropy_with_logits
    loss = input - input * target + max_val + ((-max_val).exp() + (-input - max_val).exp()).log()
RuntimeError: Expected object of type torch.cuda.FloatTensor but found type torch.cuda.LongTensor for argument #2 'other'

I encountered this problem while training your model on a customized dataset, would you please help me with that.
I followed every step as you said but still not working. Have been trying this for 2 days, would really appreciate it if you could help me.

Hi, as title,

I downloaded the Ottawa-dataset here but it looks like this structure:

How can I make it to follow the Roadnet dataset structure defined here?

Should I manually create this folder structure?
Thank you so much!

Thank you for using your task.

But I have a question about testing the DeepCrack.sh
Training the DeepCrack file is good, but I've faced the issue about python 3.5.5 maybe.

---------- Networks initialized -------------
[Network G] Total number of parameters : 14.720 M

Exception in thread Thread-1:
Traceback (most recent call last):
File "/opt/anaconda3/envs/deepsegmentor3/lib/python3.5/threading.py", line 914, in _bootstrap_inner
self.run()
File "/opt/anaconda3/envs/deepsegmentor3/lib/python3.5/threading.py", line 862, in run
self._target(*self._args, **self._kwargs)
File "/opt/anaconda3/envs/deepsegmentor3/lib/python3.5/multiprocessing/resource_sharer.py", line 139, in _serve
signal.pthread_sigmask(signal.SIG_BLOCK, range(1, signal.NSIG))
File "/opt/anaconda3/envs/deepsegmentor3/lib/python3.5/signal.py", line 60, in pthread_sigmask
sigs_set = _signal.pthread_sigmask(how, mask)
ValueError: signal number 32 out of range

Upper one is that I faced the issue.
Already I fixed the num_workers=0 but test_deepcrack.sh wasn't working.
What can I do for that issue?
Please check that issue, Thanks for your favor.

You've put together a fantastic resource here, @yhlleo. Thank you for that.

I'm attempting to perform a test run of RoadNet using the pretrained model you've provided on your dataset, but keep encountering the same error message. When I run sh ./scripts/test_roadnet.sh 0 I get the following print out:

usage: test.py [-h] --dataroot DATAROOT [--name NAME] [--gpu_ids GPU_IDS]
               [--checkpoints_dir CHECKPOINTS_DIR] [--model MODEL]
               [--input_nc INPUT_NC] [--output_nc OUTPUT_NC] [--ngf NGF]
               [--ndf NDF] [--netD NETD] [--netG NETG]
               [--n_layers_D N_LAYERS_D] [--norm NORM] [--init_type INIT_TYPE]
               [--init_gain INIT_GAIN] [--no_dropout]
               [--dataset_mode DATASET_MODE] [--direction DIRECTION]
               [--serial_batches] [--num_threads NUM_THREADS]
               [--batch_size BATCH_SIZE] [--load_size LOAD_SIZE]
               [--crop_size CROP_SIZE] [--max_dataset_size MAX_DATASET_SIZE]
               [--preprocess PREPROCESS] [--no_flip NO_FLIP]
               [--display_winsize DISPLAY_WINSIZE] [--epoch EPOCH]
               [--load_iter LOAD_ITER] [--verbose] [--suffix SUFFIX]
               [--load_width LOAD_WIDTH] [--load_height LOAD_HEIGHT]
               [--use_augment] [--display_sides DISPLAY_SIDES]
               [--num_classes NUM_CLASSES] [--use_selu USE_SELU]
               [--loss_mode LOSS_MODE] [--ntest NTEST]
               [--results_dir RESULTS_DIR] [--aspect_ratio ASPECT_RATIO]
               [--phase PHASE] [--eval] [--num_test NUM_TEST]
test.py: error: unrecognized arguments: 0

As I'm relatively inexperienced when it comes to a project of this scale, and I was unable to figure out why the message was occurring myself, I thought I'd get in touch. I'm wondering if it could be an issue with my set up as I'm working on Windows and have been unable to use conda, working from my own virtual environment instead.

How many epochs did you train the deepcrack on your dataset for fine tuning and obtaining same results of the paper?

At the end of Step #5:

RuntimeError: Unexpected error from cudaGetDeviceCount(). Did you run some cuda functions before calling NumCudaDevices() that might have already set an error? Error 803: system has unsupported display driver / cuda driver combination

Nice project!

Please update your evaluation methods ASAP so that I can compare it with other segmentation methods easily, thanks!

The environment.yml file not updated, and has the exploding gradient issue.

I changed to python 3.8 and Pytorch 2.0.1, and the training ran smoothly. One other thing need to update is the tensor self.label needs to be converted to float before feed into self.criterionSeg, in file deepcrack_model.py.

When will the Fused+CRF+GF version be uploaded?

Thank you for your work and code. I run the code according to the method you provided, but my result is much lower than yours.
There are some differences between the parameters in the train_deepcrack.sh file and those in your paper, such as learning rate decay. Maybe this is what caused the lower results when I train the Deepcrack-BN model.
So could you provide the opt.txt file of your Deepcrack-BN and Deepcrack-Aug models? I am sorry to trouble you, thank you again.

can we test on images which do not have a ground truth and have no need to calculate accuracy

Hello,

I'm trying to run your code using the provided Google Colab notebook but upon issuing the command sudo sh scripts/test_deepcrack.sh 0; I get the following error:

----------------- Options ---------------
             aspect_ratio: 1.0                           
               batch_size: 1                             
          checkpoints_dir: ./checkpoints                 
                crop_size: 256                           
                 dataroot: ./datasets/DeepCrack            [default: None]
             dataset_mode: deepcrack                     
                direction: AtoB                          
            display_sides: 1                             
          display_winsize: 256                           
                    epoch: latest                        
                     eval: False                         
                  gpu_ids: 0                             
                init_gain: 0.02                          
                init_type: xavier                        
                 input_nc: 3                             
                  isTrain: False                           [default: None]
             lambda_fused: 1.0                           
              lambda_side: 1.0                           
              load_height: 0                             
                load_iter: 0                               [default: 0]
                load_size: 256                           
               load_width: 0                             
                loss_mode: focal                         
         max_dataset_size: inf                           
                    model: deepcrack                       [default: test]
               n_layers_D: 3                             
                     name: deepcrack                       [default: experiment_name]
                      ndf: 64                            
                     netD: basic                         
                     netG: resnet_9blocks                
                      ngf: 64                            
               no_dropout: False                         
                  no_flip: 1                             
                     norm: batch                           [default: instance]
                    ntest: inf                           
              num_classes: 1                               [default: 2]
                 num_test: 10000                           [default: 50]
              num_threads: 2                             
                output_nc: 3                             
                    phase: test                          
               preprocess: resize_and_crop               
              results_dir: ./results/                    
           serial_batches: False                         
                   suffix:                               
              use_augment: False                         
                 use_selu: 1                             
                  verbose: False                         
----------------- End -------------------
dataset [DeepCrackDataset] was created
Traceback (most recent call last):
  File "test.py", line 19, in <module>
    model = create_model(opt)      # create a model given opt.model and other options
  File "/content/DeepSegmentor/models/__init__.py", line 65, in create_model
    instance = model(opt)
  File "/content/DeepSegmentor/models/deepcrack_model.py", line 45, in init
    self.gpu_ids)
  File "/content/DeepSegmentor/models/deepcrack_networks.py", line 90, in define_deepcrack
    return init_net(net, init_type, init_gain, gpu_ids)
  File "/content/DeepSegmentor/models/networks.py", line 106, in init_net
    net.to(gpu_ids[0])
  File "/usr/local/lib/python3.7/dist-packages/torch/nn/modules/module.py", line 899, in to
    return self._apply(convert)
  File "/usr/local/lib/python3.7/dist-packages/torch/nn/modules/module.py", line 570, in _apply
    module._apply(fn)
  File "/usr/local/lib/python3.7/dist-packages/torch/nn/modules/module.py", line 570, in _apply
    module._apply(fn)
  File "/usr/local/lib/python3.7/dist-packages/torch/nn/modules/module.py", line 593, in _apply
    param_applied = fn(param)
File "/usr/local/lib/python3.7/dist-packages/torch/nn/modules/module.py", line 897, in convert
    return t.to(device, dtype if t.is_floating_point() or t.is_complex() else None, non_blocking)
RuntimeError: CUDA error: all CUDA-capable devices are busy or unavailable
CUDA kernel errors might be asynchronously reported at some other API call,so the stacktrace below might be incorrect.
For debugging consider passing CUDA_LAUNCH_BLOCKING=1.

Do you have any idea on what is going on? I also tried restarting the Colab instance and verifying the GPU status with nvidia-smi but everything looks fine (output below).

+-----------------------------------------------------------------------------+
| NVIDIA-SMI 495.44       Driver Version: 460.32.03    CUDA Version: 11.2     |
|-------------------------------+----------------------+----------------------+
| GPU  Name        Persistence-M| Bus-Id        Disp.A | Volatile Uncorr. ECC |
| Fan  Temp  Perf  Pwr:Usage/Cap|         Memory-Usage | GPU-Util  Compute M. |
|                               |                      |               MIG M. |
|===============================+======================+======================|
|   0  Tesla K80           Off  | 00000000:00:04.0 Off |                    0 |
| N/A   66C    P8    30W / 149W |      0MiB / 11441MiB |      0%   E. Process |
|                               |                      |                  N/A |
+-------------------------------+----------------------+----------------------+
                                                                               
+-----------------------------------------------------------------------------+
| Processes:                                                                  |
|  GPU   GI   CI        PID   Type   Process name                  GPU Memory |
|        ID   ID                                                   Usage      |
|=============================================================================|
|  No running processes found                                                 |
+-----------------------------------------------------------------------------+

Let me know if there is any additional information I can provide to help solving the issue.

Thank you for your time and have a nice day.

thank you for your work and code
i download your code and run the program(without afterprocess including gf and crf). i gain image results and evalues them. but metrics are lower than those of your paper.

what should i do to improve the results. could you give me some advises?
thank you again

Hi, I am trying to use this code to get the crak of a picture that is taken by myself.
I followed the steps described in Readme and I use sh ./scripts/train_deepcrack.sh <gpu_id> for training.
Then I would like to do Crack Detection by executing sh ./scripts/test_deepcrack.sh <gpu_id>.
From my perspective I should add the new testing pictures to datasets/DeepCrack/test_img and by executing sh ./scripts/test_deepcrack.sh <gpu_id> I would get the identified crack images in datasets/DeepCrack/test_lab, however, the program not only requires the images in test_img folder, but also requires images in test_lab, which I thought would be the identified results.
I appreciate if anyone could explain this question to me.

Why do you set the BATCH_SIZE = 1, I think it will make the training really slow.

_C:\dl\Repository\Others\DeepSegmentor-master>conda env create -f environment.yml

Collecting package metadata (repodata.json): done
Solving environment: failed

ResolvePackageNotFound:

• pytorch=0.4.1_

Hello,
I am trying to implement the model for a custom dataset but I am having the aforementioned issue. Although I have pytorch=1.1.0 in my own virtual env, even then I am getting an assertion error as below:

(myenv) C:\dl\Repository\Others\DeepSegmentor-master>python ./tools/calculate_weights.py --data_path C:\dl\Repository\Others\DeepSegmentor-master\tools
{}
Traceback (most recent call last):
File "./tools/calculate_weights.py", line 60, in
print(reverse_weight(weights))
File "./tools/calculate_weights.py", line 43, in reverse_weight
assert len(w) > 0, "Expected a non-empty weight dict."
AssertionError: Expected a non-empty weight dict.

What should I do to solve this?