Input needs to be 8-bit YUV.
clp = core.cld.ComputeCLD(src)CLD matrices are saved as row-major ordered lists in the frame properties CLD_y, CLD_u and CLD_v.
Example snippet for finding the frame in a clip that best matches some screenshot, computing the difference as per the spec:
cld_clip = core.cld.ComputeCLD(src)
sc = core.cld.ComputeCLD(screenshot_clip)
sc_cld = {i:v for i,v in sc.et_frame(0).props.items() if "CLD" in i}
diffs = []
for k,f in enumerate(cld_clip.frames()):
cld = {i:v for i,v in f.props.items() if "CLD" in i}
diff = 0
for channel, dct in cld.items():
ch_diff = 0
for i,v in enumerate(dct):
ch_diff += (v - sc_cld[channel][i])**2
diff += math.sqrt(ch_diff)
diffs.append(diff)
print(diffs.index(min(diffs))