Remake and recap comparison

[Anonymous]

Hi all, 

I was wondering if anyone might be able to help me explain the differences in two models I have produced.

The photos I have used are exactly the same in each model and the quality of the Recap models is much higher is some places i.e around the door and the inclusion of the pub sign on the side. Is anyone able to explain why this is? I was on the understanding that both pieces of software would produce the same model but clearly, this isn't the case. I was wondering if one is designed specifically for large buildings (ReCap) where the other is designed for smaller objects (Remake). Does anyone know if they use different algorithms?

I am doing this comparison for my dissertation so if anyone can point me in the right direction for literature about the two pieces of software that would be great also.

Thank you in advance!

[Anonymous]

Hi @Anonymous,

as far as i know the engine is the very same. However, could you provide the settings you use in Remake and Recap (smart texture on/off, smart cropping on/off). Also, take in consideration that the process of alignment, point cloud creation and geometry reconstruction is not going to give you the very same result all the time, there will surely be an high degree of confidence, but outcomes could differ.

Finally, for better comparison, please provide renders one can overlap (as much as you can) so it is easier to catch the difference. I would also suggestyou to run a difference analysis of the two model, so you can do a finer evaluation of the discrepancy/consistency of your results.

Cheers,

V.

[Anonymous]

Hi @Anonymous

Both models have the smart cropping on, the recap model has the nadir optimisation on and the smart texture is on for the remake model.

Also I am not sure what you mean by renders which overlap, do you mean two models in the same image, so saving the separate images in a different file format?

Many thanks,

Amy

[Anonymous]

Hi @Anonymous,

"Both models have the smart cropping on, the recap model has the nadir optimisation on and the smart texture is on for the remake model."

When comparing result is always better to reduce variables. Try to compare the models by using the very same settings.

"Also I am not sure what you mean by renders which overlap, do you mean two models in the same image, so saving the separate images in a different file format?"

What i meant is to export the two images having (as much as feasible) the very same orientation and distance from the camera, so one can immediately and finely appreciate the difference. Also, it is always advisable to show mesh wireframe as well, as it would be easier to understand whether the differences is related to the geometry reconstruction, the texture mapping or both.

Cheers,

V.

[Anonymous]

Hi @Anonymous

Thanks for that, I have done as you have suggested and I can see that the models are much similar, therefore suggesting that the settings are what is affecting the final models, I have attached the comparisons below. This doesn't explain why the models turn out different each time but this is something I need to research. 

Are the mesh images good enough to see what the differences are related too?

Thanks

[Anonymous · Accepted solution]

Hi @Anonymous,

as i told the process of reconstruction relapses on a process involving probabilistic interpretation of the data, that is why results may differ. However, strong consistency (of course within a range of error) can be achieved by properly and finely collecting your dataset. Based on what i can see your dataset have a little amount of proper informations to feed the algorithm, so to interpolate the data Remake engine has to guess the missing data. This artificial creation of data becomes heavily variable when data interval is bigger, as in your case.

Large environment are really challenging, and even more challenging is the process to develop a system for consistent scanning results.

Start experimenting with a smaller object with a complex shape of unknown size. Scan one object with a known size 10-15 times and measure the outcomes. Do stats, record results. Then, scan another object with unknown size an equal amount of times. By using reference points set scale & measure the digital reconstructions. Do stats, record results, compare with the previous results. Evaluate margins of errors. Eventually, you will end by having a strong set of data to predict what conditions you may need to have a specific result within a known range of error.

I hope this may help!

Cheers,

V.

[Anonymous]

Hi @Anonymous

Thank you for that, I have done some modelling with smaller elements but they have not come out perfectly either so I will keep try to improve the data set and seeing where I get.

Thank you!