Host nodes: DepthMerger. #163
Conversation
kkeroo
requested review from
tersekmatija,
jkbmrz and
klemen1999
as code owners
Codecov ReportAttention: Patch coverage is
✅ All tests successful. No failed tests found.
Additional details and impacted files@@ Coverage Diff @@
## main #163 +/- ##
==========================================
- Coverage 35.88% 34.94% -0.95%
==========================================
Files 70 72 +2
Lines 4013 4124 +111
==========================================
+ Hits 1440 1441 +1
- Misses 2573 2683 +110 ☔ View full report in Codecov by Sentry. |
There was a problem hiding this comment.
I would add from .depth_merger import DepthMerger.
|
@klemen1999 Also I just remembered we were talking about tests. Should they be implemented now in those PRs that add nodes? |
There was a problem hiding this comment.
Generally LGTM
Agree on moving nodes into separate folder (same level as ml) and not nesting them into subfolders at least for now, we can do that later on if needed.
@klemen1999 Also I just remembered we were talking about tests. Should they be implemented now in those PRs that add nodes?
Good catch, we should make tests for every new node as potentially many different experiments will rely on their specifics. We can sync offline what/how to test.
| ] | ||
| ) | ||
|
|
||
| self.shrinking_factor = 0 |
There was a problem hiding this comment.
Make this also configurable through init
| @@ -0,0 +1,171 @@ | |||
| # HostSpatialsCalc implementation taken from here: | |||
There was a problem hiding this comment.
This comment can be removed
| self.depth_alignment_socket = depth_alignment_socket | ||
|
|
||
| # Values | ||
| self.DELTA = 5 # Take 10x10 depth pixels around point for depth averaging |
There was a problem hiding this comment.
Let's make this 3 values also configurable through init with default values
This PR adds first host node called DepthMerger. The node merges normal
ImgDetectionsorImgDetectionsExtendedwith depth information intoSpatialImgDetections. The node is taken from experiments host-nodes folder and slightly adjusted so it works withImgDetectionsExtended. We also introduce parametershrinking_factorthat is responsible for optionally shrinking the ROI for calculating depth values. This is useful because sometimes bounding-box also includes background and so the mean depth value is affected. E.g. In bouding-box of a human the shrinking factor will shrink the ROI into the human torso.I am not sure about the folder hierarchy. I was thinking of placing it in the
helpersfolder under specific subfolders. Maybe there is something better?The proposed node works with rewritten experiments: Human-Machine safety and social distancing.