opencv/modules/gapi/scripts/measure_privacy_masking.py

100 lines
3.5 KiB
Python
Raw Normal View History

#!/usr/bin/env python3
import sys
import subprocess
import re
from enum import Enum
## Helper functions ##################################################
##
def fmt_bool(x):
return ("true" if x else "false")
def fmt_bin(base, prec, model):
return "%s/%s/%s/%s.xml" % (base, model, prec, model)
## The script itself #################################################
##
if len(sys.argv) != 3:
print("Usage: %s /path/to/input/video /path/to/models" % sys.argv[0])
exit(1)
input_file_path = sys.argv[1]
intel_models_path = sys.argv[2]
app = "bin/example_gapi_privacy_masking_camera"
intel_fd_model = "face-detection-retail-0005"
intel_lpd_model = "vehicle-license-plate-detection-barrier-0106"
output_file = "out_results.csv"
tgts = [ ("CPU", "INT8")
, ("CPU", "FP32")
, ("GPU", "FP16")
]
class Policy(Enum):
Traditional = 1
Streaming = 2
# From mode to cmd arg
mods = [ (Policy.Traditional, True)
, (Policy.Streaming, False)
]
class UI(Enum):
With = 1
Without = 2
# From mode to cmd arg
ui = [ (UI.With, False)
, (UI.Without, True)
]
fd_fmt_bin = lambda prec : fmt_bin(intel_models_path, prec, intel_fd_model)
lpd_fmt_bin = lambda prec : fmt_bin(intel_models_path, prec, intel_lpd_model)
# Performance comparison table
table={}
# Collect the performance data
for m in mods: # Execution mode (trad/stream)
for u in ui: # UI mode (on/off)
for f in tgts: # FD model
for p in tgts: # LPD model
cmd = [ app
, ("--input=%s" % input_file_path) # input file
, ("--faced=%s" % f[0]) # FD device target
, ("--facem=%s" % fd_fmt_bin(f[1])) # FD model @ precision
, ("--platd=%s" % p[0]) # LPD device target
, ("--platm=%s" % lpd_fmt_bin(p[1])) # LPD model @ precision
, ("--trad=%s" % fmt_bool(m[1])) # Execution policy
, ("--noshow=%s" % fmt_bool(u[1])) # UI mode (show/no show)
]
out = str(subprocess.check_output(cmd))
match = re.search('Processed [0-9]+ frames \(([0-9]+\.[0-9]+) FPS\)', out)
fps = float(match.group(1))
print(cmd, fps, "FPS")
table[m[0],u[0],f,p] = fps
# Write the performance summary
# Columns: all other components (mode, ui)
with open(output_file, 'w') as csv:
# CSV header
csv.write("FD,LPD,Serial(UI),Serial(no-UI),Streaming(UI),Streaming(no-UI),Effect(UI),Effect(no-UI)\n")
for f in tgts: # FD model
for p in tgts: # LPD model
row = "%s/%s,%s/%s" % (f[0], f[1], p[0], p[1]) # FD precision, LPD precision
row += ",%f" % table[Policy.Traditional,UI.With, f,p] # Serial/UI
row += ",%f" % table[Policy.Traditional,UI.Without,f,p] # Serial/no UI
row += ",%f" % table[Policy.Streaming, UI.With, f,p] # Streaming/UI
row += ",%f" % table[Policy.Streaming, UI.Without,f,p] # Streaming/no UI
effect_ui = table[Policy.Streaming,UI.With, f,p] / table[Policy.Traditional,UI.With, f,p]
effect_noui = table[Policy.Streaming,UI.Without,f,p] / table[Policy.Traditional,UI.Without,f,p]
row += ",%f,%f" % (effect_ui,effect_noui)
row += "\n"
csv.write(row)
print("DONE: ", output_file)