Hello, everyone.
I’m working on an interesting project and I’m pretty confused, so I think I need to have a little chat on it. So please feel free to state whatever comes to mind.
Question: 2 Is there a way to tell if the object is really similar? (Visually)
PS: Type or parameter, length, distance to an object, not as the starting and ending point. Completely visually.
Sound like you want to create a ViewTemplate with filters to me
Actually, it’s worth thinking simple. I’m talking about simpliness, as understanding without parameters that 2 spheres are really spheres.
Lunchbox package Machine Learning nodes maybe
Edit: I also Topologic package might help. Shapes which are similar share the same topology.
are you talking about actual images? Or geometry? Or Images of geometry?
visually comparing things is hard - geometrically comparing them is also hard.
I think generally, tessellating stuff and then comparing the vert positions and numbers is a good starting point if you have model data and lots of time and you’re pretty sure your tessellator will always return the same results for the same input.
If you care about visual similarity then you can compare pixel by pixel with some fuzz factor - probably some ML similarity metrics for images is also an OK approach for this … myself - I would prefer doing histogram comparison or something that I can actually debug.
update:
maybe you’ll find this interesting -
Dynamo/HelixImageComparisonTests.cs at master · DynamoDS/Dynamo (github.com)
we do image comparisons of dynamo’s 3d background preview as part of our test harness - it requires a fuzz factor, and it’s not great - it misses small changes, and is very sensitive to zoom level of the image - but it catches large regressions.
@Durmus_Cesur, you should take a look to Mask RCNN(MASK RCNN / Github_mask_rcnn
Thank you for your answer, I’ve worked this and I’ve had very different results. Dynamo is very skilled at this and is improving every day. I figured that out months ago. I will share with you all I have achieved in the future.
Thank you for your concern. Machine Learning is really great.