Files
platform-external-webrtc/webrtc/video/full_stack_plot.py
ivica 092b13384e Collecting encode_time_ms for each frame.
Also, in Sample struct, replacing double with the original type.
It makes more sense to save the original data as truthful as possible, and then
convert it to double later if necessary (in the plot script).

Review URL: https://codereview.webrtc.org/1374233002

Cr-Commit-Position: refs/heads/master@{#10184}
2015-10-06 14:13:50 +00:00

415 lines
13 KiB
Python
Executable File

#!/usr/bin/env python
# Copyright (c) 2015 The WebRTC project authors. All Rights Reserved.
#
# Use of this source code is governed by a BSD-style license
# that can be found in the LICENSE file in the root of the source
# tree. An additional intellectual property rights grant can be found
# in the file PATENTS. All contributing project authors may
# be found in the AUTHORS file in the root of the source tree.
"""Generate graphs for data generated by loopback tests.
Usage examples:
Show end to end time for a single full stack test.
./full_stack_plot.py -df end_to_end -o 600 --frames 1000 vp9_data.txt
Show simultaneously PSNR and encoded frame size for two different runs of
full stack test. Averaged over a cycle of 200 frames. Used e.g. for
screenshare slide test.
./full_stack_plot.py -c 200 -df psnr -drf encoded_frame_size \\
before.txt after.txt
Similar to the previous test, but multiple graphs.
./full_stack_plot.py -c 200 -df psnr vp8.txt vp9.txt --next \\
-c 200 -df sender_time vp8.txt vp9.txt --next \\
-c 200 -df end_to_end vp8.txt vp9.txt
"""
import argparse
from collections import defaultdict
import itertools
import sys
import matplotlib.pyplot as plt
import numpy
# Fields
DROPPED = 0
INPUT_TIME = 1 # ms (timestamp)
SEND_TIME = 2 # ms (timestamp)
RECV_TIME = 3 # ms (timestamp)
RENDER_TIME = 4 # ms (timestamp)
ENCODED_FRAME_SIZE = 5 # bytes
PSNR = 6
SSIM = 7
ENCODE_TIME = 8 # ms (time interval)
TOTAL_RAW_FIELDS = 9
SENDER_TIME = TOTAL_RAW_FIELDS + 0
RECEIVER_TIME = TOTAL_RAW_FIELDS + 1
END_TO_END = TOTAL_RAW_FIELDS + 2
RENDERED_DELTA = TOTAL_RAW_FIELDS + 3
FIELD_MASK = 255
# Options
HIDE_DROPPED = 256
RIGHT_Y_AXIS = 512
# internal field id, field name, title
_fields = [
# Raw
(DROPPED, "dropped", "dropped"),
(INPUT_TIME, "input_time_ms", "input time"),
(SEND_TIME, "send_time_ms", "send time"),
(RECV_TIME, "recv_time_ms", "recv time"),
(ENCODED_FRAME_SIZE, "encoded_frame_size", "encoded frame size"),
(PSNR, "psnr", "PSNR"),
(SSIM, "ssim", "SSIM"),
(RENDER_TIME, "render_time_ms", "render time"),
(ENCODE_TIME, "encode_time_ms", "encode time"),
# Auto-generated
(SENDER_TIME, "sender_time", "sender time"),
(RECEIVER_TIME, "receiver_time", "receiver time"),
(END_TO_END, "end_to_end", "end to end"),
(RENDERED_DELTA, "rendered_delta", "rendered delta"),
]
name_to_id = {field[1]: field[0] for field in _fields}
id_to_title = {field[0]: field[2] for field in _fields}
def field_arg_to_id(arg):
if arg == "none":
return None
if arg in name_to_id:
return name_to_id[arg]
if arg + "_ms" in name_to_id:
return name_to_id[arg + "_ms"]
raise Exception("Unrecognized field name \"{}\"".format(arg))
class PlotLine(object):
"""Data for a single graph line."""
def __init__(self, label, values, flags):
self.label = label
self.values = values
self.flags = flags
class Data(object):
"""Object representing one full stack test."""
def __init__(self, filename):
self.title = ""
self.length = 0
self.samples = defaultdict(list)
self._read_samples(filename)
def _read_samples(self, filename):
"""Reads graph data from the given file."""
f = open(filename)
it = iter(f)
self.title = it.next().strip()
self.length = int(it.next())
field_names = [name.strip() for name in it.next().split()]
field_ids = [name_to_id[name] for name in field_names]
for field_id in field_ids:
self.samples[field_id] = [0.0] * self.length
for sample_id in xrange(self.length):
for col, value in enumerate(it.next().split()):
self.samples[field_ids[col]][sample_id] = float(value)
self._subtract_first_input_time()
self._generate_additional_data()
f.close()
def _subtract_first_input_time(self):
offset = self.samples[INPUT_TIME][0]
for field in [INPUT_TIME, SEND_TIME, RECV_TIME, RENDER_TIME]:
if field in self.samples:
self.samples[field] = [x - offset for x in self.samples[field]]
def _generate_additional_data(self):
"""Calculates sender time, receiver time etc. from the raw data."""
s = self.samples
last_render_time = 0
for field_id in [SENDER_TIME, RECEIVER_TIME, END_TO_END, RENDERED_DELTA]:
s[field_id] = [0] * self.length
for k in range(self.length):
s[SENDER_TIME][k] = s[SEND_TIME][k] - s[INPUT_TIME][k]
decoded_time = s[RENDER_TIME][k]
s[RECEIVER_TIME][k] = decoded_time - s[RECV_TIME][k]
s[END_TO_END][k] = decoded_time - s[INPUT_TIME][k]
if not s[DROPPED][k]:
if k > 0:
s[RENDERED_DELTA][k] = decoded_time - last_render_time
last_render_time = decoded_time
def _hide(self, values):
"""
Replaces values for dropped frames with None.
These values are then skipped by the plot() method.
"""
return [None if self.samples[DROPPED][k] else values[k]
for k in range(len(values))]
def add_samples(self, config, target_lines_list):
"""Creates graph lines from the current data set with given config."""
for field in config.fields:
# field is None means the user wants just to skip the color.
if field is None:
target_lines_list.append(None)
continue
field_id = field & FIELD_MASK
values = self.samples[field_id]
if field & HIDE_DROPPED:
values = self._hide(values)
target_lines_list.append(PlotLine(
self.title + " " + id_to_title[field_id],
values, field & ~FIELD_MASK))
def average_over_cycle(values, length):
"""
Returns the list:
[
avg(values[0], values[length], ...),
avg(values[1], values[length + 1], ...),
...
avg(values[length - 1], values[2 * length - 1], ...),
]
Skips None values when calculating the average value.
"""
total = [0.0] * length
count = [0] * length
for k in range(len(values)):
if values[k] is not None:
total[k % length] += values[k]
count[k % length] += 1
result = [0.0] * length
for k in range(length):
result[k] = total[k] / count[k] if count[k] else None
return result
class PlotConfig(object):
"""Object representing a single graph."""
def __init__(self, fields, data_list, cycle_length=None, frames=None,
offset=0, output_filename=None, title="Graph"):
self.fields = fields
self.data_list = data_list
self.cycle_length = cycle_length
self.frames = frames
self.offset = offset
self.output_filename = output_filename
self.title = title
def plot(self, ax1):
lines = []
for data in self.data_list:
if not data:
# Add None lines to skip the colors.
lines.extend([None] * len(self.fields))
else:
data.add_samples(self, lines)
def _slice_values(values):
if self.offset:
values = values[self.offset:]
if self.frames:
values = values[:self.frames]
return values
length = None
for line in lines:
if line is None:
continue
line.values = _slice_values(line.values)
if self.cycle_length:
line.values = average_over_cycle(line.values, self.cycle_length)
if length is None:
length = len(line.values)
elif length != len(line.values):
raise Exception("All arrays should have the same length!")
ax1.set_xlabel("Frame", fontsize="large")
if any(line.flags & RIGHT_Y_AXIS for line in lines if line):
ax2 = ax1.twinx()
ax2.set_xlabel("Frame", fontsize="large")
else:
ax2 = None
# Have to implement color_cycle manually, due to two scales in a graph.
color_cycle = ["b", "r", "g", "c", "m", "y", "k"]
color_iter = itertools.cycle(color_cycle)
for line in lines:
if not line:
color_iter.next()
continue
if self.cycle_length:
x = numpy.array(range(self.cycle_length))
else:
x = numpy.array(range(self.offset, self.offset + len(line.values)))
y = numpy.array(line.values)
ax = ax2 if line.flags & RIGHT_Y_AXIS else ax1
ax.plot(x, y, "o-", label=line.label, markersize=3.0, linewidth=1.0,
color=color_iter.next())
ax1.grid(True)
if ax2:
ax1.legend(loc="upper left", shadow=True, fontsize="large")
ax2.legend(loc="upper right", shadow=True, fontsize="large")
else:
ax1.legend(loc="best", shadow=True, fontsize="large")
def load_files(filenames):
result = []
for filename in filenames:
if filename in load_files.cache:
result.append(load_files.cache[filename])
else:
data = Data(filename)
load_files.cache[filename] = data
result.append(data)
return result
load_files.cache = {}
def get_parser():
class CustomAction(argparse.Action):
def __call__(self, parser, namespace, values, option_string=None):
if "ordered_args" not in namespace:
namespace.ordered_args = []
namespace.ordered_args.append((self.dest, values))
parser = argparse.ArgumentParser(
description=__doc__, formatter_class=argparse.RawDescriptionHelpFormatter)
parser.add_argument(
"-c", "--cycle_length", nargs=1, action=CustomAction,
type=int, help="Cycle length over which to average the values.")
parser.add_argument(
"-f", "--field", nargs=1, action=CustomAction,
help="Name of the field to show. Use 'none' to skip a color.")
parser.add_argument("-r", "--right", nargs=0, action=CustomAction,
help="Use right Y axis for given field.")
parser.add_argument("-d", "--drop", nargs=0, action=CustomAction,
help="Hide values for dropped frames.")
parser.add_argument("-o", "--offset", nargs=1, action=CustomAction, type=int,
help="Frame offset.")
parser.add_argument("-n", "--next", nargs=0, action=CustomAction,
help="Separator for multiple graphs.")
parser.add_argument(
"--frames", nargs=1, action=CustomAction, type=int,
help="Frame count to show or take into account while averaging.")
parser.add_argument("-t", "--title", nargs=1, action=CustomAction,
help="Title of the graph.")
parser.add_argument(
"-O", "--output_filename", nargs=1, action=CustomAction,
help="Use to save the graph into a file. "
"Otherwise, a window will be shown.")
parser.add_argument(
"files", nargs="+", action=CustomAction,
help="List of text-based files generated by loopback tests.")
return parser
def _plot_config_from_args(args, graph_num):
# Pylint complains about using kwargs, so have to do it this way.
cycle_length = None
frames = None
offset = 0
output_filename = None
title = "Graph"
fields = []
files = []
mask = 0
for key, values in args:
if key == "cycle_length":
cycle_length = values[0]
elif key == "frames":
frames = values[0]
elif key == "offset":
offset = values[0]
elif key == "output_filename":
output_filename = values[0]
elif key == "title":
title = values[0]
elif key == "drop":
mask |= HIDE_DROPPED
elif key == "right":
mask |= RIGHT_Y_AXIS
elif key == "field":
field_id = field_arg_to_id(values[0])
fields.append(field_id | mask if field_id is not None else None)
mask = 0 # Reset mask after the field argument.
elif key == "files":
files.extend(values)
if not files:
raise Exception("Missing file argument(s) for graph #{}".format(graph_num))
if not fields:
raise Exception("Missing field argument(s) for graph #{}".format(graph_num))
return PlotConfig(fields, load_files(files), cycle_length=cycle_length,
frames=frames, offset=offset, output_filename=output_filename,
title=title)
def plot_configs_from_args(args):
"""Generates plot configs for given command line arguments."""
# The way it works:
# First we detect separators -n/--next and split arguments into groups, one
# for each plot. For each group, we partially parse it with
# argparse.ArgumentParser, modified to remember the order of arguments.
# Then we traverse the argument list and fill the PlotConfig.
args = itertools.groupby(args, lambda x: x in ["-n", "--next"])
args = list(list(group) for match, group in args if not match)
parser = get_parser()
plot_configs = []
for index, raw_args in enumerate(args):
graph_args = parser.parse_args(raw_args).ordered_args
plot_configs.append(_plot_config_from_args(graph_args, index))
return plot_configs
def show_or_save_plots(plot_configs):
for config in plot_configs:
fig = plt.figure(figsize=(14.0, 10.0))
ax = fig.add_subplot(1, 1, 1)
plt.title(config.title)
config.plot(ax)
if config.output_filename:
print "Saving to", config.output_filename
fig.savefig(config.output_filename)
plt.close(fig)
plt.show()
if __name__ == "__main__":
show_or_save_plots(plot_configs_from_args(sys.argv[1:]))