Generate ground truth offsets for my own dataset

[HongChow]

Is this repo design both for semantic segmentation and for segfix ?
What about if I only want to train segfix?
As the model zoo says:
"
For whom want to try SegFix by training a new SegFix model, you should:

1. Generate ground truth offsets.
2. Download ImageNet pretrained model to pretrained_model/.
3. Run the training script.
4. Run the prediction script to predict offsets for Cityscapes val / test set.
5. Run the refinement script to refine any labels with the offsets.
   "
   and I want to know how to generate ground truth offsets.

[PkuRainBow]

@hsfzxjy Please help to illustrate the guidelines to generate the ground-truth offsets.

[HongChow]

@hsfzxjy Please help to illustrate the guidelines to generate the ground-truth offsets.

thank you ，I have already done this. by changing some code in lib/datasets/preprocess/cityscapes/dt_offset_generator.py

[xiamibudayang]

@HongChow Did you meet the error "NotImplementedError: Got 3D input, but bilinear mode needs 4D input
" in "openseg.pytorch-master/lib/loss/loss_helper.py" when you train the segfix model by your own datasets?

[140ai]

@HongChow Do you solve it? How to apply it?

[jhyin12]

It is appreciated if you tell me how to generate ground truth offsets in your own dataset?

[sym330]

It is appreciated if you tell me how to generate ground truth offsets in your own dataset?