Recherche avancée

Sur d’autres sites (6793)

• ffmpeg : how to extract keyframes from certain time ranges into filenames with timestamps from the original video ?

  18 mai 2023, par Display Name

  I have as input specific timestamps I'm interested in, and I wish to extract keyframes closest to them. I thus use skip_frame nokey and a select='between(t,...,...)+between(t,...,...)+...' where I add a few seconds around each time I'm interested in (enough so that at least one keyframe will fall in that range based on the input video I have, and then can manually delete if more than one came out in a given time range in my sequence). Chaining the between()s lets me use a single command to extract all of these images in order to avoid seeking from the beginning of the video for each image, were I to use separate command per image. So this part works fine.

  


  The problem is that I want the output image filenames to correspond to the timestamps, in seconds (or some decimal fraction of seconds like tenths or milliseconds) of the extracted frames with respect to the INPUT video. With older versions of ffmpeg, I used to be able for example to get output filenames being times in terms of tenths of a second with -vsync 0 -r 10 -frame_pts true %05d.webp but with recent versions, that results in the error One of -r/-fpsmax was specified together a non-CFR -vsync/-fps_mode. This is contradictory. Replacing the deprecated -vsync with -fps_mode and one of the CFR values results in ffmpeg DUPLICATING frames to fulfill the specified -r value which results in a huge number of output images. The way I am able to get just the set of keyframes I want and no duplications is to drop the -r and use -fps_mode passthrough, but then I lose the naming of the output files by their time in the original video. Searching around here and elsewhere on the web, I have tried things like setting settb=...,setpts=... and -copyts but in the end couldn't get it to work.

  


  As a complete example, the command
ffmpeg -skip_frame nokey -i "input.mp4" -vf "select='between(t,15,25)+between(t,40,50)+between(t,95,105)+between(t,120,130)+between(t,190,200)',scale='min(480,iw)':-2:flags=lanczos" -fps_mode passthrough -copyts -c:v libwebp -quality 75 -preset photo -frame_pts true %05d.webp gives me the right set of output images, but not the filenames that would make it easy for me to quickly manually find frames corresponding to specific times in the original video.

  


• ffmpeg extract keyframes from certain time ranges to filenames with timestamps from original video

  17 mai 2023, par Display Name

  I have as input specific timestamps I'm interested in, and I wish to extract keyframes closest to them. I thus use skip_frame nokey and a select='between(t,...,...)+between(t,...,...)+...' where I add a few seconds around each time I'm interested in (enough so that at least one keyframe will fall in that range based on the input video I have, and then can manually delete if more than one came out in a given time range in my sequence). Chaining the between()s lets me use a single command to extract all of these images in order to avoid seeking from the beginning of the video for each image, were I to use separate command per image. So this part works fine.

  


  The problem is that I want the output image filenames to correspond to the timestamps, in seconds (or some decimal fraction of seconds like tenths or milliseconds) of the extracted frames with respect to the INPUT video. With older versions of ffmpeg, I used to be able for example to get output filenames being times in terms of tenths of a second with -vsync 0 -r 10 -frame_pts true %05d.webp but with recent versions, that results in the error One of -r/-fpsmax was specified together a non-CFR -vsync/-fps_mode. This is contradictory. Replacing the deprecated -vsync with -fps_mode and one of the CFR values results in ffmpeg DUPLICATING frames to fulfill the specified -r value which results in a huge number of output images. The way I am able to get just the set of keyframes I want and no duplications is to drop the -r and use -fps_mode passthrough, but then I lose the naming of the output files by their time in the original video. Searching around here and elsewhere on the web, I have tried things like setting settb=...,setpts=... and -copyts but in the end couldn't get it to work.

  


  As a complete example, the command
ffmpeg -skip_frame nokey -i "input.mp4" -vf "select='between(t,15,25)+between(t,40,50)+between(t,95,105)+between(t,120,130)+between(t,190,200)',scale='min(480,iw)':-2:flags=lanczos" -fps_mode passthrough -copyts -c:v libwebp -quality 75 -preset photo -frame_pts true %05d.webp gives me the right set of output images, but not the filenames that would make it easy for me to quickly manually find frames corresponding to specific times in the original video.

  


• Why Yuv data readed from ffmpeg is different from original input yuv ?

  13 mai 2021, par sianyi Huang

  I use ffmpeg to make HDR test video, my approach is write a image, converting the image to yuv420p and then use ffmpeg to make the HDR test video.

  



  But I found the yuv data readed from mp4 is different from the original input..
I was stucked in here for a while, does anyone know how to read the correct yuv data from mp4 ?

  



  #ffmpeg encode command
ffmpeg_encode_mp4 = \
"ffmpeg -y -s 100*100 -pix_fmt yuv420p -threads 4 -r 1 -stream_loop -1 -f rawvideo -i write_yuv.yuv -vf \
scale=out_h_chr_pos=0:out_v_chr_pos=0,format=yuv420p10le \
-c:v libx265 -tag:v hvc1 -t 10 -pix_fmt yuv420p10le -preset medium -x265-params \
crf=12:colorprim=bt2020:transfer=smpte2084:colormatrix=bt2020nc:master-display=\"G(13250,34500)B(7500,3000)R(34000,16000)WP(15635,16450)L(10000000,1)\":max-cll=\"1000,400\" \
-an test.mp4"

#ffmpeg read yuv from mp4 command
ffmpeg_extract_yuv = "ffmpeg -i test.mp4 -vframes 1 -c:v rawvideo -pix_fmt yuv420p read_yuv.yuv"


#make 100*100 yuv raw
w, h = 100, 100
test_gray = 255
test = np.full((100, 100, 3), test_gray, dtype=np.uint8)
yuv_cv = cv.cvtColor(test, cv.COLOR_RGB2YUV_I420)
yuv_cv.tofile("write_yuv.yuv")

#encode yuv raw to mp4 with HDR metadata
print(ffmpeg_encode_mp4)
result = subprocess.check_output(ffmpeg_encode_mp4, shell = True)
print(result)
sleep(0.5)

#extract yuv from mp4
kill_existing_file("read_yuv.yuv")
print(ffmpeg_extract_yuv)
result = subprocess.check_output(ffmpeg_extract_yuv, shell = True)
print(result)
sleep(0.5)

write_yuv = np.fromfile("write_yuv.yuv",dtype='uint8')
read_yuv = np.fromfile("read_yuv.yuv",dtype='uint8')

print("input gray:", test_gray)
print("write_yuv", write_yuv[:10])
print("read_yuv", read_yuv[:10])

reader = imageio.get_reader("test.mp4")
img = reader.get_data(0)
print("imgeio read:", img[50, 50])

'''
ouput result:
input gray: 255
write_yuv [235 235 235 235 235 235 235 235 235 235]
read_yuv [234 235 234 235 234 235 234 235 234 235]
imgeio read: [253 253 253]
'''


  



  I have no idea how to validate the video I made is corret
Any feedback will be very appreciated !