Day 73: Interpreting parsing maps for measurement extraction
Turning regions into measurements
A region's raw extent isn't a measurement yet — you want anatomically meaningful widths. Chest width is best taken as the torso region's width at a specific *height* (roughly armpit level), not its maximum width (which might catch an arm). This is where parsing, pose, and silhouette fuse: use the pose shoulders to find the right vertical position, then measure the torso region's width there.
def chest_width_px(parsing_map, shoulder_y, label=UPPER_CLOTHES):
# measure the torso region's width at (just below) shoulder height
row = parsing_map[int(shoulder_y) + 10] # a bit below shoulders
xs = np.where(row == label)[0]
if len(xs) == 0:
return None
return {"px": int(xs.max() - xs.min()), "source": "parsing+pose"}Fusion resolves single-source failures
Pose alone can misplace a shoulder; parsing alone can't tell chest height from hem height. Together — pose locates *where*, parsing bounds *what* — each covers the other's blind spot. When your Day-84 tape-measure validation shows a dimension is off, you'll have multiple sources to compare and diagnose *which* one failed, instead of one opaque number.
Key terms
- Anatomical measurement
- A width/length taken at a meaningful body location (e.g. chest at armpit level), not just a region's max extent.
- Pose-parsing fusion
- Using pose to locate the vertical position and parsing to bound the region, for a targeted measurement.
Why measure chest width as the torso region width at shoulder height, rather than the torso region's maximum width?
