x and y coordinates are swapped

[ktugan]

x and y coordinates are swapped

[ktugan]

how to:

# copies in a static manner the values of a framecontainer into a new one
def deepcopy_framecontainer(fc, fc2):
    fc2.camId = fc.camId
    
    fc2.init('dataSources', len(fc.dataSources))
    for i, datasource in enumerate(fc.dataSources):
        fc2.dataSources[i] = fc.dataSources[i]
    fc2.fromTimestamp = fc.fromTimestamp
    fc2.hiveId = fc.hiveId
    fc2.id = fc.id
    fc2.toTimestamp = fc.toTimestamp
    
    fc2.init('transformationMatrix', len(fc.transformationMatrix))
    for i, tm in enumerate(fc.transformationMatrix):
        fc2.transformationMatrix[i] = fc.transformationMatrix[i]
    
    fc2.init('frames', len(fc.frames))
    for i, frame in enumerate(fc.frames):
        fc2.frames[i] = fc.frames[i]


from bb_binary import Repository, FrameContainer

repo = Repository('data/1h_09-25-15/')
repo_fixed = Repository('data/1h_09-25-15_fixed2/')

for fname in repo.iter_fnames():
    fc = load_frame_container(fname)
    fc2 = FrameContainer.new_message()
    deepcopy_framecontainer(fc, fc2)
    
    for frame in fc2.frames:
    	for detection in frame.detectionsUnion.detectionsDP:
    		detection.x, detection.y = detection.y, detection.x
    
    repo_fixed.add(fc2)

[gitmirgut]

If all DetectionDP.xpos and DetectionDP.ypos in the whole repro are swapped, i think it is faster to just change the bb_binary_schema.capnp.
From:

struct DetectionDP {
...
  xpos @1 :UInt16;               # x coordinate of the grid center wrt. the image
  ypos @2 :UInt16;               # y coordinate of the grid center wrt. the image
...
}

To:

struct DetectionDP {
...
  ypos @1 :UInt16;               # x coordinate of the grid center wrt. the image
  xpos @2 :UInt16;               # y coordinate of the grid center wrt. the image
...
}

But then we have to fix the output of DP simultaneous, otherwise we produce multiple inconsistent data, where xpos and ypos are sometimes swapped and sometimes not.

If I am not mistaken, the swap of the coordinates in the DP was happening in bb_pipeline-processing.py at line L107.

positions = peak_local_max(im, min_distance=dist)

peak_local_max returns the row (y-Axis) first and then the column(x-Axis).

[nebw]

See also BioroboticsLab/bb_pipeline@c8f8942

Thanks for the suggestion @gitmirgut