search

bertjiazheng / Structured3D

[ECCV'20] Structured3D: A Large Photo-realistic Dataset for Structured 3D Modeling
https://structured3d-dataset.org
MIT License
541 stars 62 forks source link

Only about 77% of bounding boxes labels can be inferred #17

Open Xia0ben opened 4 years ago

Xia0ben commented 4 years ago

Hello,

I followed your advice in Issue #13 to infer the semantic labels associated with the bounding boxes, but have only succeeded in inferring 77% of them from the instance.png and semantic.png images of each scene (for 438311 bounding boxes, only 337345 could be inferred, that is more exactly 76.9647579002%).

This missing quarter of information is quite big, and seems as I feared in my message in Issue #13, to be due to object occlusion in the images. Therefore, it seems that it is impossible for end users to infer the missing data. Would you please consider filling in the gap =) ?

Wish you a nice day !

bertjiazheng commented 4 years ago

I am currently involved in another project. I will try to add this feature in my spare time.

Xia0ben commented 4 years ago

Thank you very much for considering it, that would be very helpful ! =)

supriya-gdptl commented 4 years ago

Hello @Xia0ben, Could you please create a pull request with the code you used to infer bounding box labels? It would be of great help! Thank you

Xia0ben commented 4 years ago

@supriya-gdptl I just uploaded the code I wrote to infer the semantic labels from the images for the bounding boxes on a fork of this repository: https://github.com/Xia0ben/Structured3D. I do not have the time to cleanup/document the code right now, so it would be nice to do a pull-request directly to @bertjiazheng 's main repo if you were to use it and clean it up a bit. I mainly reused existing code written by @bertjiazheng , there is not much more to it but it would indeed be helpful if a proper script were added to the main repo. ;-)

Cheers, Benoit.