Part-in -whole matching

[bnjff]

Thanks for your work on this library!
I got good results matching a model to individual objects extracted from the point cloud, however so far had no luck fitting the model to a more cluttered scene. Since the relevant part of the wiki is not available yet, could you please give a brief guidelines to make part-in-whole matching work?

[nmellado]

Hi @bnjff
Yes, the part-in-whole documentation is still not available.

The main reason is that I could find robust-enough general guidelines working on multiple examples.
This is not a limitation of the registration technique itself, but rather a limitation of the current implementation, which is too much tailored for the pairwise registration case.

After some time thinking about this problem, I believe the Super4PCS application needs a kind of expert mode, providing more parameters to the user. Presets could be added for the part-in-whole case.
The main improvement would be to provide more control on the sampling and base extraction stages.
I believe we need:

• a basis defined in the object cloud, with its size controlled wrt the object size
• same for the overlap, this parameter should be expressed wrt the object size
• an automatic estimation of the sampling parameters of the scene, wrt to the base size, etc.

This clearly needs a dedicated development effort.
I can try to put this idea in the next release (after 1.1.3).

[bnjff]

To work with the current implementation, I ended up running object detection on RGB and using that to crop out the ROI point cloud with the object.

[nmellado]

Thanks for the feedback @bnjff .
Could you share more details about you approach ? Is it fully automated ?Is it something we could add to the library ?

[aigerd]

guys, note that if the dataset in which you want to find an object is given offline and you have enoght space (e.g. O(n^2)), there is no point in using the (super)4PCS. in this case, one can index that data such that all the needed computations are done offline.

…

On Mon, Nov 13, 2017 at 6:20 PM, Nicolas Mellado ***@***.***> wrote: Thanks for the feddback @bnjff <https://github.com/bnjff> . Could you share more details about you approach ? Is it fully automated ?Is it something we could add to the library ? — You are receiving this because you are subscribed to this thread. Reply to this email directly, view it on GitHub <#46 (comment)>, or mute the thread <https://github.com/notifications/unsubscribe-auth/AGYw95j5naKk8WXSlDjav28Kr4uf0kG2ks5s2GxOgaJpZM4PIB5C> .

[nmellado]

Thanks @aigerd , that's definitively true.
BTW, it might be interesting to add such brute force approach in the library for offline processing and GT generation.

[Giessen]

Hi, is the "Part-in -whole matching" supported or not? I tried to match part of a point clound with another but the result is random... tks

[Giessen]

I would like to match object shown in this picture. Even with 2000 points, the result is still random. Any suggestions, please?

[Giessen]

Obj files attached.
Hex.zip