How to prepare the Cityscapes data

[arash-ashra]

Hello. I'm trying to reproduce your CityScapes results for our BMVC paper.

after I followed the data directory format in the config.profile file and running bash ./scripts/cityscapes/hrnet/run_h_48_d_4_ocr.sh val 1 I get this error:

ERROR: Found no prediction for ground truth /home/arash/openseg.pytorch/dataset/cityscapes/val/label/munster_000027_000019_gtFine_labelIds.png

could you explain how did you prepare the data?
Thanks

[PkuRainBow]

According to the error information, it seems that you have not downloaded the prediction files of the "HRNet + OCR" method.

As for the data preparation, please refer to the GETTING_STARTED.

[arash-ashra]

where can I find these prediction files of the HRNet + OCR method? and why it is needed?
This is my full traceback:
Traceback (most recent call last): File "main.py", line 212, in <module> model = Tester(configer) File "/home/arash/openseg.pytorch/segmentor/tester.py", line 69, in __init__ self._init_model() File "/home/arash/openseg.pytorch/segmentor/tester.py", line 72, in _init_model self.seg_net = self.model_manager.semantic_segmentor() File "/home/arash/openseg.pytorch/lib/models/model_manager.py", line 81, in semantic_segmentor model = SEG_MODEL_DICT[model_name](self.configer) File "/home/arash/openseg.pytorch/lib/models/nets/hrnet.py", line 105, in __init__ self.backbone = BackboneSelector(configer).get_backbone() File "/home/arash/openseg.pytorch/lib/models/backbones/backbone_selector.py", line 34, in get_backbone model = HRNetBackbone(self.configer)(**params) File "/home/arash/openseg.pytorch/lib/models/backbones/hrnet/hrnet_backbone.py", line 598, in __call__ bn_momentum=0.1) File "/home/arash/openseg.pytorch/lib/models/backbones/hrnet/hrnet_backbone.py", line 307, in __init__ self.bn1 = ModuleHelper.BatchNorm2d(bn_type=bn_type)(64, momentum=bn_momentum) TypeError: 'NoneType' object is not callable evaluate the result... ERROR: Found no prediction for ground truth /home/arash/openseg.pytorch/dataset/cityscapes/val/label/munster_000113_000019_leftImg8bit.png

[PkuRainBow]

bash ./scripts/cityscapes/hrnet/run_h_48_d_4_ocr.sh val 1

The above command is used to evaluate the performance of HRNet + OCR, so you need to generate the predictions of HRNet + OCR by yourself in the first place. Please download the pre-trained checkpoints provided in the MODEL_ZOO.md

[fabio-cancio-sena]

bash ./scripts/cityscapes/hrnet/run_h_48_d_4_ocr.sh val 1

The above command is used to evaluate the performance of HRNet + OCR, so you need to generate the predictions of HRNet + OCR by yourself in the first place. Please download the pre-trained checkpoints provided in the MODEL_ZOO.md

Hello,

I've been reading through the getting started, model zoo and some scripts and I'm still a little confused about the workflow to run segfix. Does segfix run HRNet + OCR during training after I convert the dataset? What are the steps and expected inputs and outputs? What are this pre-trained checkpoints? Are they pretrained HRNet + OCR (for example) or HRNet + OCR + SegFix or just SegFix?

[PkuRainBow]

SegFix is independent of the segmentation models and you can apply any segmentation models to predict the coarse segmentation maps.

SegFix predicts an offset map and we can apply the offset map to refine the coarse segmentation maps.

Hope the above explanation helps you!

[AxMM]

bash ./scripts/cityscapes/hrnet/run_h_48_d_4_ocr.sh val 1

The above command is used to evaluate the performance of HRNet + OCR, so you need to generate the predictions of HRNet + OCR by yourself in the first place. Please download the pre-trained checkpoints provided in the MODEL_ZOO.md

Hello.
I´m getting the same error as @arashash
I did the Data Preparation step for cityscapes.

• My save_dir is openseg.pytorch-master/dataroot/cityscapes,
• my ori_root_dir is openseg.pytorch-master/dataroot/cityscapes/original_cityscapes with 2 folders from the cityscapes website the folders are: gtCoarse (with train, train_extra and val) and leftImg8bit. Inside leftImg8bit I have: test, train, train_extra and val

I downloaded the checkpoint (hrnet_w48_ocr_1_latest.pth) then I put it in the folder openseg.pytorch-master/checkpoints/cityscapes/

I downloaded ImageNet pretrained model to openseg.pytorch-master/pretrained_model/:

• hrnetv2_w48_imagenet_pretrained.pth
• hr_rnet_bt_w20_imagenet_pretrained.pth

I´m getting the error ERROR: Found no prediction for ground truth /home/mysusername/openseg.pytorch-master/dataroot/cityscapes/val/label/munster_000119_000019_leftImg8bit.png

What I´m missing? can you help me?
Thank you in advance

[PkuRainBow]

@hsfzxjy Please help @AxMM to address the problem.

[hsfzxjy]

@AxMM

Seems that the script didn't perform inference. Could you provide the full output of bash ./scripts/cityscapes/hrnet/run_h_48_d_4_ocr.sh val 1?

[AxMM]

@AxMM

Seems that the script didn't perform inference. Could you provide the full output of bash ./scripts/cityscapes/hrnet/run_h_48_d_4_ocr.sh val 1?

Hello,
Thank you for the fast reply.
I figure it out. My problem is because I'm using Windows 10. I tried to solve it by installing WSL (Windows Subsystem for Linux) with that I got the error that I'm reporting.
There is any chance to put your project working on Windows?

[hsfzxjy]

@AxMM It's not on the plan. Adapting Windows will be too time-consuming and troublesome. Instead Linux is more mature and stable.

[AxMM]

@AxMM It's not on the plan. Adapting Windows will be too time-consuming and troublesome. Instead Linux is more mature and stable.

I understand, thank you very much. I will switch for Linux.

[AxMM]

Hello,
Running bash ./scripts/cityscapes/hrnet/run_h_48_d_4_ocr.sh val 1 I got this error:
FileNotFoundError: [Errno 2] No such file or directory: '/msravcshare/dataset/cityscapes/val/image' evaluate the result... ERROR: Found no prediction for ground truth /dl/openseg.pytorch-master/DATA_ROOT/cityscapes/val/label/lindau_000011_000019_leftImg8bit.png

I don´t recognize this folder msravcshare.
What is the purpose of this folder?
My DATA_ROOT is DATA_ROOT=/dl/openseg.pytorch-master/DATA_ROOT

Thank you in advance.

[hsfzxjy]

@AxMM Fixed. You may checkout the latest commit to see the update.

[xiximelon]

where can I find these prediction files of the HRNet + OCR method? and why it is needed? This is my full traceback: Traceback (most recent call last): File "main.py", line 212, in <module> model = Tester(configer) File "/home/arash/openseg.pytorch/segmentor/tester.py", line 69, in __init__ self._init_model() File "/home/arash/openseg.pytorch/segmentor/tester.py", line 72, in _init_model self.seg_net = self.model_manager.semantic_segmentor() File "/home/arash/openseg.pytorch/lib/models/model_manager.py", line 81, in semantic_segmentor model = SEG_MODEL_DICT[model_name](self.configer) File "/home/arash/openseg.pytorch/lib/models/nets/hrnet.py", line 105, in __init__ self.backbone = BackboneSelector(configer).get_backbone() File "/home/arash/openseg.pytorch/lib/models/backbones/backbone_selector.py", line 34, in get_backbone model = HRNetBackbone(self.configer)(**params) File "/home/arash/openseg.pytorch/lib/models/backbones/hrnet/hrnet_backbone.py", line 598, in __call__ bn_momentum=0.1) File "/home/arash/openseg.pytorch/lib/models/backbones/hrnet/hrnet_backbone.py", line 307, in __init__ self.bn1 = ModuleHelper.BatchNorm2d(bn_type=bn_type)(64, momentum=bn_momentum) TypeError: 'NoneType' object is not callable evaluate the result... ERROR: Found no prediction for ground truth /home/arash/openseg.pytorch/dataset/cityscapes/val/label/munster_000113_000019_leftImg8bit.png

@arashash Hello .I met the same question.Could u tell me how to solve the problem? I follow the pipeline the author provided. Seems I can't get the prediction of models. How can I get the predictions? Thanks!

[xiximelon]

bash ./scripts/cityscapes/hrnet/run_h_48_d_4_ocr.sh val 1

The above command is used to evaluate the performance of HRNet + OCR, so you need to generate the predictions of HRNet + OCR by yourself in the first place. Please download the pre-trained checkpoints provided in the MODEL_ZOO.md

@PkuRainBow Hello I met the same question as the qustioner. You say "you need to generate the predictions of HRNet + OCR by yourself". Does this mean the code can't perform the inference operation? According to the GETTING_STARTED. I think it means by running the command bash ./scripts/cityscapes/hrnet/run_h_48_d_4_ocr.sh val 1 ,it can do the inference.But now I met the error ERROR: Found no prediction for ground truth .Please help me . Thanks a lot!