Recherche avancée

Sur d’autres sites (5575)

• Trying to crop an image in ffmpeg using alphamerge but produces wrong alpha

  12 septembre 2021, par Alex Styl

  I am using alphamerge in order to crop an image within a circle.

  


  What I have so far is :

  


  ffmpeg -f lavfi -i "color=c=white:size=240x240" -i avatar.png -i mask.png -filter_complex \
 "[1][2]alphamerge[img]; \
 [0][img]overlay[out]" -c:v png -map "[out]" -pix_fmt rgba -t 5 -y out.mp4 2>&1


  


  with avatar.png and mask.png respectively being :

  


  

  


  This produces the following output (1 frame of the output video) :

  


  

  


  which is unexpected, given the original input is much darker than this.

  


  How can I crop the 'avatar.png' using the 'mask.png' so that the output is the avatar.png cropped in a circle and keeping the same alpha ?

  


  PS : The important bit here is for me to be able to crop the original image and maintaining the correct colors/apha of the original image. If there is an other way of doing this (other than alphamerge) I am happy to hear it.

  


• How to retrieve, process and display frames from a capture device with minimal latency

  14 mars 2024, par valle

  I'm currently working on a project where I need to retrieve frames from a capture device, process them, and display them with minimal latency and compression. Initially, my goal is to maintain the video stream as close to the source signal as possible, ensuring no noticeable compression or latency. However, as the project progresses, I also want to adjust framerate and apply image compression.

  


  I have experimented using FFmpeg, since that was the first thing that came to my mind when thinking about capturing video(frames) and processing them.

  


  However I am not satisfied yet, since I am experiencing delay in the stream. (No huge delay but definately noticable)
The command that worked best so far for me :

  


  ffmpeg -rtbufsize 512M -f dshow -i video="Blackmagic WDM Capture (4)" -vf format=yuv420p -c:v libx264 -preset ultrafast -qp 0 -an -tune zerolatency -f h264 - | ffplay -fflags nobuffer -flags low_delay -probesize 32 -sync ext -

  


  I also used OBS to capture the video stream from the capture device and when looking into the preview there was no noticable delay. I then tried to simulate the exact same settings using ffmpeg :

  


  ffmpeg -rtbufsize 512M -f dshow -i video="Blackmagic WDM Capture (4)" -vf format=yuv420p -r 60 -c:v libx264 -preset veryfast -b:v 2500K -an -tune zerolatency -f h264 - | ffplay -fflags nobuffer -flags low_delay -probesize 32 -sync ext -

  


  But the delay was kind of similar to the one of the command above.
I know that OBS probably has a lot complexer stuff going on (Hardware optimization etc.) but atleast I know this way that it´s somehow possible to display the stream from the capture device without any noticable latency (On my setup).

  


  The approach that so far worked best for me (In terms of delay) was to use Python and OpenCV to read frames of the capture device and display them. I also implemented my own framerate (Not perfect I know) but when it comes to compression I am rather limited compared to FFmpeg and the frame processing is also too slow when reaching framerates about 20 fps and more.

  


  import cv2
import time

# Set desired parameters
FRAME_RATE = 15  # Framerate in frames per second
COMPRESSION_QUALITY = 25  # Compression quality for JPEG format (0-100)
COMPRESSION_FLAG = True   # Enable / Disable compression

# Set capture device index (replace 0 with the index of your capture card)
cap = cv2.VideoCapture(4, cv2.CAP_DSHOW)

# Check if the capture device is opened successfully
if not cap.isOpened():
    print("Error: Could not open capture device")
    exit()

# Create an OpenCV window
# TODO: The window is scaled to fullscreen here (The source video is 1920x1080, the display is 1920x1200)
#       I don´t know the scaling algorithm behind this, but it seems to be a simple stretch / nearest neighbor
cv2.namedWindow('Frame', cv2.WINDOW_NORMAL)
cv2.setWindowProperty('Frame', cv2.WND_PROP_FULLSCREEN, cv2.WINDOW_FULLSCREEN)

# Loop to capture and display frames
while True:
    # Start timer for each frame processing cycle
    start_time = time.time()

    # Capture frame-by-frame
    ret, frame = cap.read()

    # If frame is read correctly, proceed
    if ret:
        if COMPRESSION_FLAG:
            # Perform compression
            _, compressed_frame = cv2.imencode('.jpg', frame, [int(cv2.IMWRITE_JPEG_QUALITY), COMPRESSION_QUALITY])
            # Decode the compressed frame
            frame = cv2.imdecode(compressed_frame, cv2.IMREAD_COLOR)

        # Display the frame
        cv2.imshow('Frame', frame)

        # Calculate elapsed time since the start of this frame processing cycle
        elapsed_time = time.time() - start_time

        # Calculate available time for next frame
        available_time = 1.0 / FRAME_RATE

        # Check if processing time exceeds available time
        if elapsed_time > available_time:
            print("Warning: Frame processing time exceeds available time.")

        # Calculate time to sleep to achieve desired frame rate -> maintain a consistent frame rate
        sleep_time = 1.0 / FRAME_RATE - elapsed_time

        # If sleep time is positive, sleep to control frame rate
        if sleep_time > 0:
            time.sleep(sleep_time)

    # Break the loop if 'q' is pressed
    if cv2.waitKey(1) & 0xFF == ord('q'):
        break

# Release the capture object and close the display window
cap.release()
cv2.destroyAllWindows()


  


  I also thought about getting the SDK of the capture device in order to upgrade the my performance.
But Since I am not used to low level programming but rather to scripting languages, I thought I would reach out to the StackOverflow community at first, and see if anybody has some hints to better approaches or any tips how I could increase my performance.

  


  Any Help is appreciated !

  


• Minimal Understanding of VP8′s Forward Transform

  16 novembre 2010, par Multimedia Mike — VP8

  Regarding my toy VP8 encoder, Pengvado mentioned in the comments of my last post, “x264 looks perfect using only i16x16 DC mode. You must be doing something wrong in computing residual or fdct or quantization.” This makes a lot of sense. The encoder generates a series of elements which describe how to reconstruct the original image. Intra block reconstruction takes into consideration the following elements :

  
  
  

  I have already verified that both my encoder and FFmpeg’s VP8 decoder agree precisely on how to reconstruct blocks based on the predictors, coefficients, and quantizers. Thus, if the decoded image still looks crazy, the elements the encoder is generating to describe the image must be wrong.

  So I started studying the forward DCT, which I had cribbed wholesale from the original libvpx 0.9.0 source code. It should be noted that the formal VP8 spec only defines the inverse transform process, not the forward process. I was using a version designated as the “short” version, vs. the “fast” version. Then I looked at the 0.9.5 FDCT. Then I got the idea of comparing the results of each.

  input:   92 91 89 86 91 90 88 86 89 89 89 88 89 87 88 93

  • libvpx 0.9.0 “short” :

    forward : -314 5 1 5 4 5 -2 0 0 1 -1 -1 1 11 -3 -4
    inverse : 92 91 89 86 89 86 91 90 91 90 88 86 88 86 89 89
    

  • libvpx 0.9.0 “fast” :

    forward : -314 4 0 5 4 4 -2 0 0 1 0 -1 1 11 -2 -5
    inverse : 91 91 89 86 88 86 91 90 91 90 88 86 88 86 89 89
    

  • libvpx 0.9.5 “short” :

    forward : -312 7 1 0 1 12 -5 2 2 -3 3 -1 1 0 -2 1
    inverse : 92 91 89 86 91 90 88 86 89 89 89 88 89 87 88 93
    

  I was surprised when I noticed that input[] != idct(fdct(input[])) in some of the above cases. Then I remembered that the aforementioned property isn’t what is meant by a “bit-exact” transform– only that all implementations of the inverse transform are supposed to produce bit-exact output for a given vector of input coefficients.

  Anyway, I tried applying each of these forward transforms. I got slightly differing results, with the latest one I tried (the fdct from libvpx 0.9.5) producing the best results (to my eye). At least the trees look better in the Big Buck Bunny logo image :

  
  
  The dense trees of the Big Buck Bunny logo using one of the libvpx 0.9.0 forward transforms
  
  
  The same segment of the image using the libvpx 0.9.5 forward transform
  

  Then again, it could be that the different numbers generated by the newer forward transform triggered different prediction modes to be chosen. Overall, adapting the newer FDCT did not dramatically improve the encoding quality.

  Working on the intra 4×4 mode encoding is generating some rather more accurate blocks than my intra 16×16 encoder. Pengvado indicated that x264 generates perfectly legible results when forcing the encoder to only use intra 16×16 mode. To be honest, I’m having trouble understanding how that can possibly occur thanks to the Walsh-Hadamard transform (WHT). I think that’s where a lot of the error is creeping in with my intra 16×16 encoder. Then again, FFmpeg implements an inverse WHT function that bears ‘vp8′ in its name. This implies that it’s custom to the algorithm and not exactly shared with H.264.