Bug? Radius Search with Autotuned Index

[felixkling]

Hi,

I want to perform a fixed radius search with FLANN using the autotuned index (to minimize the search time). My problem involves a large number of point (about a million) at a dimension around 5 - 10 resulting to a number of nearest neighbors in the range 0-10000. I need to find (approximately) all nearest neighbors within a fixed radius. If I use the Autotuned Index, it only gives me the 5-10 nearest neighbors.

I think i found the problem: In "autotuned_index.h" the "bestSearchParams_" are used in the function "radiusSearch". When calculating the best search parameters in "estimateSearchParams" also the variable "checks" is evaluated and set to a relatively low number.

Is there an autotuned function available for the radius search that returns all nearest neighbors within the radius?

Felix

[andersgb1]

Maybe I'm wrong here, but I thought that in any case, if your data has
dimension <= 10, then it's always best to use a standard kd-tree (
http://nl.mathworks.com/help/stats/createns.html), which also happens to
produce exact results.

On 18 November 2014 00:15, felixkling notifications@github.com wrote:

Hi,

I want to perform a fixed radius search with FLANN using the autotuned
index (to minimize the search time). My problem involves a large number of
point (about a million) at a dimension around 5 - 10 resulting to a number
of nearest neighbors in the range 0-10000. I need to find (approximately)
all nearest neighbors within a fixed radius. If I use the Autotuned Index,
it only gives me the 5-10 nearest neighbors.

I think i found the problem: In "autotuned_index.h" the
"bestSearchParams_" are used in the function "radiusSearch". When
calculating the best search parameters in "estimateSearchParams" also the
variable "checks" is evaluated and set to a relatively low number.

Is there an autotuned function available for the radius search that
returns all nearest neighbors within the radius?

Felix

—
Reply to this email directly or view it on GitHub
#219.

[tpet]

This may be related. I just noticed that one has to use radius squared. Unfortunately, this is not mentioned in the docs. It is nevertheless consistent with FLANN also returning distances squared.