Mismatch Between GOOSE 2D RGB Images and Color Segmentation

[probstlukas]

Hello,

First of all, thank you for providing and supporting the GOOSE dataset!

While working with the GOOSE 2D Images validation split, I noticed a mismatch between the RGB images and color segmentation labels. So far, this issue appears to be isolated to the place 2023-05-17_neubiberg_sunny.

In several scenes, the following issues were observed:

• *_color.png does not align spatially with *_windshield_vis.png
• In some cases, labels appear to be from different time frames or scenes

---

Example Scene with Mismatch

Scene: 2023-05-17_neubiberg_sunny__0381_1684329746496937615 (click to expand)

RGB image (windshield_vis)

Semantic segmentation (color)

Label IDs (labelids)

Instance IDs (instanceids)

In this example, the color (semantic segmentation) image does not match the windshield_vis image. The labelids and instanceids appear consistent with the RGB scene.

---

Affected Scenes

Here are other scene IDs from 2023-05-17_neubiberg_sunny where mismatches were detected:

• 2023-05-17_neubiberg_sunny__0381_1684329746496937615
• 2023-05-17_neubiberg_sunny__0383_1684329748563080364
• 2023-05-17_neubiberg_sunny__0384_1684329750435328885
• 2023-05-17_neubiberg_sunny__0391_1684329792820931736
• 2023-05-17_neubiberg_sunny__0423_1684329888720615790
• 2023-05-17_neubiberg_sunny__0426_1684329911503929495
• 2023-05-17_neubiberg_sunny__0427_1684329917847527420
• 2023-05-17_neubiberg_sunny__0428_1684329918963923280
• 2023-05-17_neubiberg_sunny__0429_1684329926681604286
• 2023-05-17_neubiberg_sunny__0430_1684329929122268538
• 2023-05-17_neubiberg_sunny__0431_1684329931747725910
• 2023-05-17_neubiberg_sunny__0432_1684329936071994519
• 2023-05-17_neubiberg_sunny__0433_1684329936179109284
• 2023-05-17_neubiberg_sunny__0434_1684329943943709425
• 2023-05-17_neubiberg_sunny__0435_1684329946225122968
• 2023-05-17_neubiberg_sunny__0436_1684329947779374894
• 2023-05-17_neubiberg_sunny__0437_1684329948193807412
• 2023-05-17_neubiberg_sunny__0438_1684329949541126973
• 2023-05-17_neubiberg_sunny__0439_1684329955810539276
• 2023-05-17_neubiberg_sunny__0440_1684329957111926151
• 2023-05-17_neubiberg_sunny__0441_1684329962187110006
• 2023-05-17_neubiberg_sunny__0442_1684329963327103198
• 2023-05-17_neubiberg_sunny__0443_1684329965034134829
• 2023-05-17_neubiberg_sunny__0444_1684329966029236668
• 2023-05-17_neubiberg_sunny__0445_1684329969028426176
• 2023-05-17_neubiberg_sunny__0446_1684329969653744873
• 2023-05-17_neubiberg_sunny__0447_1684329970581964440
• 2023-05-17_neubiberg_sunny__0448_1684329971937901805
• 2023-05-17_neubiberg_sunny__0449_1684329986241684618
• 2023-05-17_neubiberg_sunny__0450_1684329988542461527
• 2023-05-17_neubiberg_sunny__0451_1684329993012176910
• 2023-05-17_neubiberg_sunny__0452_1684329994421309306
• 2023-05-17_neubiberg_sunny__0453_1684329996494574879
• 2023-05-17_neubiberg_sunny__0455_1684330001468592718

---

If this is indeed a dataset error, would it be possible to recheck the *_color.png alignment for the listed scenes?

Please let me know if I can help further or provide more examples.

Thank you again for making this dataset available!

[pemo-ml]

Hi @probstlukas,

thank you for creating this issue!
I can confirm the mismatch of the RGB color maps (*_color.png) with the grayscale label maps (*_labelids.png) and grayscale instance maps (*_instanceids.png).

As far as I can tell, the grayscale mappings are consistent with the windshield images, so please resort to them instead of the RGB maps until we upload an updated version of the GOOSE 2D data.

I'll comment this issue when the most recent version of the GOOSE 2D data is online.
I'll leave the issue open until then.

Cheers,
Peter