Recherche avancée

Sur d’autres sites (11298)

• avfilter/f_ebur128 : lift sample peak calculation out of main loop

  13 juin, par Niklas Haas

  avfilter/f_ebur128 : lift sample peak calculation out of main loop
  This is substantially faster ( 55%) than the transposed loop, and also
  
  avoids an unnecessary macro.
  

  • [DH] libavfilter/f_ebur128.c

• How to embed additional metadata to a video file in libavcodec (FFmpeg C source code fork)

  7 mars 2023, par leavittx

  I'm extending an existing codec (Hap) inside ffmpeg library, adding some additional stuff to it (modifying FFmpeg code in my own fork of it)

  


  One of the things I'd like to do is to store a per-file maximum encoded packet size in some global header/metadata so I can use it later while doing decoding (generally, the metadata/info to store can be arbitrary, size is just what I currently need for some specialized decoder I'm writing).

  


  I'm modifying the ffmpeg code file
https://github.com/FFmpeg/FFmpeg/blob/master/libavcodec/hapenc.c

  


  EDIT : What I'd like to achieve

  


  

  • Compute some numeric value during the encoding (namely, the maximum encoded frame size for the video - it's needed for a specialized fast decoder to not read all the frames beforehand) - that is done already

  


  • Store that value in the video file metadata (looks like it should be done on container level)

  


  


  My initial incorrect attempt :
But I don't currently see any possibility to save any info outside of a single packet being produced by hap_encode function and written to AVPacket *pkt parameter.

  


  

  1. Set dict parameter on AVCodecContext *avctx's priv_data :

  


  


      av_opt_set_int(avctx->priv_data, "max_compressed_frame_size", <size>, 0);&#xA;    av_opt_set(avctx->priv_data, "max_compressed_frame_size", "", 0);&#xA;</size>

  


  Later, while decoding av_opt_get/av_opt_get_int report non-existing key.

  


  

  2. Set various AVCodecContext *avctx fields :

  


  


      avctx->flags2 = <size>;&#xA;    avctx->extradata_size = <size>;&#xA;    avctx->ticks_per_frame = <size>;&#xA;    avctx->delay = <size>;&#xA;</size></size></size></size>

  


  They all also seem to be not preserved when doing decoding later (I get just default values from decoder's AVCodecContext)

  


• Trying to get the current FPS and Frametime value into Matplotlib title

  16 juin 2022, par TiSoBr

  I try to turn an exported CSV with benchmark logs into an animated graph. Works so far, but I can't get the Titles on top of both plots with their current FPS and frametime in ms values animated.

  


  Thats the output I'm getting. Looks like he simply stores all values in there instead of updating them ?

  


  Screengrab of cli output
Screengrab of the final output (inverted)

  


  from __future__ import division
import sys, getopt
import time
import matplotlib
import numpy as np
import subprocess
import math
import re
import argparse
import os
import glob

import matplotlib.animation as animation
import matplotlib.pyplot as plt


def check_pos(arg):
    ivalue = int(arg)
    if ivalue <= 0:
        raise argparse.ArgumentTypeError("%s Not a valid positive integer value" % arg)
    return True
    
def moving_average(x, w):
    return np.convolve(x, np.ones(w), 'valid') / w
    

parser = argparse.ArgumentParser(
    description = "Example Usage python frame_scan.py -i mangohud -c '#fff' -o mymov",
    formatter_class=argparse.ArgumentDefaultsHelpFormatter)
parser.add_argument("-i", "--input", help = "Input data set from mangohud", required = True, nargs='+', type=argparse.FileType('r'), default=sys.stdin)
parser.add_argument("-o", "--output", help = "Output file name", required = True, type=str, default = "")
parser.add_argument("-r", "--framerate", help = "Set the desired framerate", required = False, type=float, default = 60)
parser.add_argument("-c", "--colors", help = "Colors for the line graphs; must be in quotes", required = True, type=str, nargs='+', default = 60)
parser.add_argument("--fpslength", help = "Configures how long the data will be shown on the FPS graph", required = False, type=float, default = 5)
parser.add_argument("--fpsthickness", help = "Changes the line width for the FPS graph", required = False, type=float, default = 3)
parser.add_argument("--frametimelength", help = "Configures how long the data will be shown on the frametime graph", required = False, type=float, default = 2.5)
parser.add_argument("--frametimethickness", help = "Changes the line width for the frametime graph", required = False, type=float, default = 1.5)
parser.add_argument("--graphcolor", help = "Changes all of the line colors on the graph; expects hex value", required = False, default = '#FFF')
parser.add_argument("--graphthicknes", help = "Changes the line width of the graph", required = False, type=float, default = 1)
parser.add_argument("-ts","--textsize", help = "Changes the the size of numbers marking the ticks", required = False, type=float, default = 23)
parser.add_argument("-fsM","--fpsmax", help = "Changes the the size of numbers marking the ticks", required = False, type=float, default = 180)
parser.add_argument("-fsm","--fpsmin", help = "Changes the the size of numbers marking the ticks", required = False, type=float, default = 0)
parser.add_argument("-fss","--fpsstep", help = "Changes the the size of numbers marking the ticks", required = False, type=float, default = 30)
parser.add_argument("-ftM","--frametimemax", help = "Changes the the size of numbers marking the ticks", required = False, type=float, default = 50)
parser.add_argument("-ftm","--frametimemin", help = "Changes the the size of numbers marking the ticks", required = False, type=float, default = 0)
parser.add_argument("-fts","--frametimestep", help = "Changes the the size of numbers marking the ticks", required = False, type=float, default = 10)

arg = parser.parse_args()
status = False


if arg.input:
    status = True
if arg.output:
    status = True
if arg.framerate:
    status = check_pos(arg.framerate)
if arg.fpslength:
    status = check_pos(arg.fpslength)
if arg.fpsthickness:
    status = check_pos(arg.fpsthickness)
if arg.frametimelength:
    status = check_pos(arg.frametimelength)
if arg.frametimethickness:
    status = check_pos(arg.frametimethickness)
if arg.colors:
    if len(arg.output) != len(arg.colors):
        for i in arg.colors:
            if re.match(r"^#([A-Fa-f0-9]{6}|[A-Fa-f0-9]{3})$", i):
                status = True
            else:
                print('{} : Isn\'t a valid hex value!'.format(i))
                status = False
    else:
        print('You must have the same amount of colors as files in input!')
        status = False
if arg.graphcolor:
    if re.match(r"^#([A-Fa-f0-9]{6}|[A-Fa-f0-9]{3})$", arg.graphcolor):
        status = True
    else:
        print('{} : Isn\'t a vaild hex value!'.format(arg.graphcolor))
        status = False
if arg.graphthicknes:
    status = check_pos(arg.graphthicknes)
if arg.textsize:
    status = check_pos(arg.textsize)
if not status:
    print("For a list of arguments try -h or --help") 
    exit()


# Empty output folder
files = glob.glob('/output/*')
for f in files:
    os.remove(f)


# We need to know the longest recording out of all inputs so we know when to stop the video
longest_data = 0

# Format the raw data into a list of tuples (fps, frame time in ms, time from start in micro seconds)
# The first three lines of our data are setup so we ignore them
data_formated = []
for li, i in enumerate(arg.input):
    t = 0
    sublist = []
    for line in i.readlines()[3:]:
        x = line[:-1].split(',')
        fps = float(x[0])
        frametime = int(x[1])/1000 # convert from microseconds to milliseconds
        elapsed = int(x[11])/1000 # convert from nanosecond to microseconds
        data = (fps, frametime, elapsed)
        sublist.append(data)
    # Compare last entry of each list with the 
    if sublist[-1][2] >= longest_data:
        longest_data = sublist[-1][2]
    data_formated.append(sublist)


max_blocksize = max(arg.fpslength, arg.frametimelength) * arg.framerate
blockSize = arg.framerate * arg.fpslength


# Get step time in microseconds
step = (1/arg.framerate) * 1000000 # 1000000 is one second in microseconds
frame_size_fps = (arg.fpslength * arg.framerate) * step
frame_size_frametime = (arg.frametimelength * arg.framerate) * step


# Total frames will have to be updated for more then one source
total_frames = int(int(longest_data) / step)


if True: # Gonna be honest, this only exists so I can collapse this block of code

    # Sets up our figures to be next to each other (horizontally) and with a ratio 3:1 to each other
    fig, (ax1, ax2) = plt.subplots(1, 2, gridspec_kw={'width_ratios': [3, 1]})

    # Size of whole output 1920x360 1080/3=360
    fig.set_size_inches(19.20, 3.6)

    # Make the background transparent
    fig.patch.set_alpha(0)


    # Loop through all active axes; saves a lot of lines in ax1.do_thing(x) ax2.do_thing(x)
    for axes in fig.axes:

        # Set all splines to the same color and width
        for loc, spine in axes.spines.items():
            axes.spines[loc].set_color(arg.graphcolor)
            axes.spines[loc].set_linewidth(arg.graphthicknes)

        # Make sure we don't render any data points as this will be our background
        axes.set_xlim(-(max_blocksize * step), 0)
        

        # Make both plots transparent as well as the background
        axes.patch.set_alpha(.5)
        axes.patch.set_color('#020202')

        # Change the Y axis info to be on the right side
        axes.yaxis.set_label_position("right")
        axes.yaxis.tick_right()

        # Add the white lines across the graphs; the location of the lines are based off set_{}ticks
        axes.grid(alpha=.8, b=True, which='both', axis='y', color=arg.graphcolor, linewidth=arg.graphthicknes)

        # Remove X axis info
        axes.set_xticks([])

    # Add a another Y axis so ticks are on both sides
    tmp_ax1 = ax1.secondary_yaxis("left")
    tmp_ax2 = ax2.secondary_yaxis("left")

    # Set both to the same values
    ax1.set_yticks(np.arange(arg.fpsmin, arg.fpsmax + 1, step=arg.fpsstep))
    ax2.set_yticks(np.arange(arg.frametimemin, arg.frametimemax + 1, step=arg.frametimestep))
    tmp_ax1.set_yticks(np.arange(arg.fpsmin , arg.fpsmax + 1, step=arg.fpsstep))
    tmp_ax2.set_yticks(np.arange(arg.frametimemin, arg.frametimemax + 1, step=arg.frametimestep))

    # Change the "ticks" to be white and correct size also change font size
    ax1.tick_params(axis='y', color=arg.graphcolor ,width=arg.graphthicknes, length=16, labelsize=arg.textsize, labelcolor=arg.graphcolor)
    ax2.tick_params(axis='y', color=arg.graphcolor ,width=arg.graphthicknes, length=16, labelsize=arg.textsize, labelcolor=arg.graphcolor)
    tmp_ax1.tick_params(axis='y', color=arg.graphcolor ,width=arg.graphthicknes, length=8, labelsize=0) # Label size of 0 disables the fps/frame numbers
    tmp_ax2.tick_params(axis='y', color=arg.graphcolor ,width=arg.graphthicknes, length=8, labelsize=0)


    # Limits Y scale
    ax1.set_ylim(arg.fpsmin,arg.fpsmax + 1)
    ax2.set_ylim(arg.frametimemin,arg.frametimemax + 1)

    # Add an empty plot
    line = ax1.plot([], lw=arg.fpsthickness)
    line2 = ax2.plot([], lw=arg.frametimethickness)

    # Sets all the data for our benchmark
    for benchmarks, color in zip(data_formated, arg.colors):
        y = moving_average([x[0] for x in benchmarks], 25)
        y2 = [x[1] for x in benchmarks]
        x = [x[2] for x in benchmarks]
        line += ax1.plot(x[12:-12],y, c=color, lw=arg.fpsthickness)
        line2 += ax2.step(x,y2, c=color, lw=arg.fpsthickness)
    
    # Add titles with values
    ax1.set_title("Avg. frames per second: {}".format(y2), color=arg.graphcolor, fontsize=20, fontweight='bold', loc='left')
    ax2.set_title("Frametime in ms: {}".format(y2), color=arg.graphcolor, fontsize=20, fontweight='bold', loc='left')  

    # Removes unwanted white space; also controls the space between the two graphs
    plt.tight_layout(pad=0, h_pad=0, w_pad=2.5)
    
    fig.canvas.draw()

    # Cache the background
    axbackground = fig.canvas.copy_from_bbox(ax1.bbox)
    ax2background = fig.canvas.copy_from_bbox(ax2.bbox)


# Create a ffmpeg instance as a subprocess we will pipe the finished frame into ffmpeg
# encoded in Apple QuickTime (qtrle) for small(ish) file size and alpha support
# There are free and opensource types that will also do this but with much larger sizes
canvas_width, canvas_height = fig.canvas.get_width_height()
outf = '{}.mov'.format(arg.output)
cmdstring = ('ffmpeg',
                '-stats', '-hide_banner', '-loglevel', 'error', # Makes ffmpeg less annoying / to much console output
                '-y', '-r', '60', # set the fps of the video
                '-s', '%dx%d' % (canvas_width, canvas_height), # size of image string
                '-pix_fmt', 'argb', # format cant be changed since this is what  `fig.canvas.tostring_argb()` outputs
                '-f', 'rawvideo',  '-i', '-', # tell ffmpeg to expect raw video from the pipe
                '-vcodec', 'qtrle', outf) # output encoding must support alpha channel
pipe = subprocess.Popen(cmdstring, stdin=subprocess.PIPE)

def render_frame(frame : int):

    # Set the bounds of the graph for each frame to render the correct data
    start = (frame * step) - frame_size_fps
    end = start + frame_size_fps
    ax1.set_xlim(start,end)
     
     
    start = (frame * step) - frame_size_frametime
    end = start + frame_size_frametime
    ax2.set_xlim(start,end)
    

    # Restore background
    fig.canvas.restore_region(axbackground)
    fig.canvas.restore_region(ax2background)

    # Redraw just the points will only draw points with in `axes.set_xlim`
    for i in line:
        ax1.draw_artist(i)
        
    for i in line2:
        ax2.draw_artist(i)

    # Fill in the axes rectangle
    fig.canvas.blit(ax1.bbox)
    fig.canvas.blit(ax2.bbox)
    
    fig.canvas.flush_events()

    # Converts the finished frame to ARGB
    string = fig.canvas.tostring_argb()
    return string




#import multiprocessing
#p = multiprocessing.Pool()
#for i, _ in enumerate(p.imap(render_frame, range(0, int(total_frames + max_blocksize))), 20):
#    pipe.stdin.write(_)
#    sys.stderr.write('\rdone {0:%}'.format(i/(total_frames + max_blocksize)))
#p.close()

#Signle Threaded not much slower then multi-threading
if __name__ == "__main__":
    for i , _ in enumerate(range(0, int(total_frames + max_blocksize))):
        render_frame(_)
        pipe.stdin.write(render_frame(_))
        sys.stderr.write('\rdone {0:%}'.format(i/(total_frames + max_blocksize)))