Видео

Hi, thanks for your interest. I haven't worked in that field for quite some years. I am not sure about COLMAP. But any structure from motion algorithm should work.

@@fanwang5609 Thank you so much for ur quick response. I have one more question: in VisualSFM do I need to check all the possible matches that have been made form each pair manually?? or there is another quicker way to do that.?? ( I'm doing some experiments for my thesis diploma for 3D reconstuction of objects with reflective and textureless surfaces and I'm using VisualSFM..I have 36 photos from each object)...Thank you again for ur time. I appreciate it :)

@@Ada1992ifyIn general, there is no quicker way to validate all pairs of matches except for doing it manually. But if you are familiar with the process, false match should be quite obvious.

@@fanwang5609 Thank you again for your quick response. I really appreciate it and helped me a lot. I have some more questions: 1) What threshold did you use? 2) Is the whole script written the file “main.cpp” (please answer with 1,2) Thank you again for ur time..:)

@@Ada1992ify 1) I am not sure as it has been a long time. 2) I don't think I understand your question.

+Bony Martinez Hi Bony, if you use our default CU-ASIFT sift detection, you may end up with huge amount of features on high resolution images. My suggestion is you find siftParam_settings.txt under the same folder with CODE.exe and change the first parameter -tilt to 5 (or 3 or 1) if your matching images do not have very wide baselines. You may even try to change the second parameter -fo to 0 which might degrade the performance. These techniques can all help you to effectively lower the feature numbers but sometimes compromise the result. You can give it a try and let me know how it goes.

+Bony Martinez Hi Bony, as the problem suggested, you need an x64 cuda 7.0 nppc dll for the program instead of cuda 7.5. You can either install cuda toolkit 7.0 to your pc or simply copy nppc64_75.dll which you should be able to locate under C:\Program Files\NVIDIA GPU Computing Toolkit\CUDA\v7.5\bin if you use the default installation path to the same folder with CODE.exe and change its name to nppc64_70.dll (this is not a standard way!!!). Let me know if this works for you.