synthia_mapped_to_cityscapes(labels)

[bjchen666]

I can not find the synthia_mapped_to_cityscapes(labels) in dataset. Who can help me?

[swamiviv]

Hi, this (code to generate the data) will be uploaded in 2-3 days.

…

On Tue, Jul 24, 2018 at 1:53 AM, gshdong ***@***.***> wrote: I can not find the synthia_mapped_to_cityscapes(labels) in dataset. Who can help me? — You are receiving this because you are subscribed to this thread. Reply to this email directly, view it on GitHub <#6>, or mute the thread <https://github.com/notifications/unsubscribe-auth/AGWIeM1BRm5gPFFFLFZPsA1BF5XMvsM1ks5uJrZDgaJpZM4VcH7v> .

-- --Swami

[bjchen666]

@swamiviv Thanks!

[bjchen666]

@swamiviv Has the code that generates the data been updated?

[YifeiAI]

@swamiviv could you please update the synthia_mapped_to_cityscapes(labels) script? I couldn't reproduce the result claimed in your paper when I use my own script for mapping the classes.

[mathfinder]

+1

[mfaraki]

Any update on this issue?

[swamiviv]

Hi, I have changed org so things have been slow. The results are very much reproducible as shown in this thread: #4

Meanwhile, the label mapping is very similar to what's being done in similar repos we have compared in the paper such as: https://github.com/YangZhang4065/AdaptationSeg?files=1

[fabioperez]

I could reproduce the paper with mIoU of 34.42.

To make the labels coherent, I used this script and changed the paths in the GT files.

I converted both SYNTHIA and CITYSCAPES labels. During evaluation, I had to delete this line since the new GT files already had the same labels.

[aserdega]

@swamiviv, could you please share the code for mapping SYNTHIA labels to cityscapes with me?
I am working on paper and wish to cite your work, however I can not reproduce the experiment without synthia_mapped_to_cityscapes labels =(
Thank you in advance!

[aserdega]

@fabioperez, Hi, do you still have your modified code for training and label mapping?

[fabioperez]

These are the files that I used: https://gist.github.com/fabioperez/0ea329e86ee5c695d1faccc806ed2d7b

The idea is to map the labels from different datasets to the same IDs. I also had to delete the following line since there the label conversion was performed offline:

LSD-seg/code/eval_cityscapes.py

Line 55 in e373e89

label = label_mapping(label, mapping)

[yingzicy]

@fabioperez, Hi, I found your script very helpful. I downloaded RAND-CITYSCAPES for synthia dataset from the website but found that images are not available, I mean, the folder only includes depth images. Any suggestion?