Closed PaulHax closed 5 months ago
Thanks for the PR, why do we need to change the return type of the object_detector to a [path:{model pred}]
as opposed to the current array of model predictions. The model keeps the order of the array that we pass as input.
object_detector.eval
still returns an array of predictions. The "image group by size/shape" code was reording the order we pass images to self.pipeline
. Then the output predictions order did not match the input paths order which made problems here:
https://github.com/Kitware/nrtk-explorer/blob/main/src/nrtk_explorer/app/transforms.py#L160-L162
object_detector.eval still returns an array of predictions. The "image group by size/shape" code was reording the order we pass images to self.pipeline. Then the output predictions order did not match the input paths order which made problems here:
You are right, we did not test for over the batching default size... Good catch here :)
Ill come back to this PR later, I think that we can further refactor.
@PaulHax I pushed a commit to this PR that adds your changes to object_dectector but also changes the output. Let me know what do you think
+1 Works good!