Recherche avancée

Sur d’autres sites (4401)

• Saving mp4 files into csv for training the data

  4 février 2021, par KSp

  I am very new to the computer vision field and I am trying to train my model and as a start of the work, I used a label encoder to label my videos for events I am using. Here I have two events which are accident and no accident.

  


  Folder Structure for the images :

  


  Colab_Notebooks
- accident(all the .jpg frames are here)
- nonaccident(all the .jpg frames are here)


  


  So my data.csv file looks like this and code given below.

  


  data.csv 
image_path,target
/content/drive/MyDrive/Colab_Notebooks/accident/accident_0000638.jpg,0.0
/content/drive/MyDrive/Colab_Notebooks/nonaccident/nonaccident_0002143.jpg,1.0
/content/drive/MyDrive/Colab_Notebooks/accident/accident_0000372.jpg,0.0
/content/drive/MyDrive/Colab_Notebooks/accident/accident_0000419.jpg,0.0
/content/drive/MyDrive/Colab_Notebooks/nonaccident/nonaccident_0001675.jpg,1.0
/content/drive/MyDrive/Colab_Notebooks/accident/accident_0000307.jpg,0.0
/content/drive/MyDrive/Colab_Notebooks/accident/accident_00001099.jpg,0.0
/content/drive/MyDrive/Colab_Notebooks/accident/accident_0000940.jpg,0.0
/content/drive/MyDrive/Colab_Notebooks/accident/accident_0000892.jpg,0.0
/content/drive/MyDrive/Colab_Notebooks/accident/accident_0000805.jpg,0.0
/content/drive/MyDrive/Colab_Notebooks/accident/accident_0000232.jpg,0.0
/content/drive/MyDrive/Colab_Notebooks/accident/accident_0000255.jpg,0.0
/content/drive/MyDrive/Colab_Notebooks/accident/accident_0000840.jpg,0.0
/content/drive/MyDrive/Colab_Notebooks/accident/accident_0000974.jpg,0.0


  


  The code i used for generating data.csv is as shown below :

  


  all_paths = os.listdir('/content/drive/MyDrive/Colab_Notebooks/')

folder_paths = [x for x in all_paths if os.path.isdir('/content/drive/MyDrive/Colab_Notebooks/' + x )]

print(f"Folder paths : {folder_paths}")
print (f"Number of folders: {len(folder_paths)}")
create_labels = ['accident','nonaccident']

data = pd.DataFrame()

image_formats = ['jpg']
labels = []
counter = 0
for i, folder_path in tqdm(enumerate(folder_paths), total = len(folder_paths)):
    if folder_path not in create_labels:
        continue
    image_paths = os.listdir('/content/drive/MyDrive/Colab_Notebooks/' + folder_path)
    label = folder_path

    for image_path in image_paths:
        if image_path.split('.')[-1] in image_formats:
            data.loc[counter,'image_path'] =  f"/content/drive/MyDrive/Colab_Notebooks/{folder_path}/{image_path}"
            labels.append(label)
            counter += 1
labels = np.array(labels)
# one-hot encode the labels
lb = LabelBinarizer()
labels = lb.fit_transform(labels)

#print(labels)

# save as CSV file
data.to_csv('/content/drive/MyDrive/Colab_Notebooks/data.csv', index=False)

# pickle the binarized labels
print('Saving the binarized labels as pickled file')
joblib.dump(lb, '/content/drive/MyDrive/Colab_Notebooks/lb.pkl')

print(data.head(5))


  


  I was able to do this fine because the dataset you see on top is frames which are jpg images. But I would like to do the same for videos.

  


  Colab_Notebooks
- accident(all the .mp4 clips are here)
- nonaccident(all the .mp4 clips are here)

Expected output:
/content/drive/MyDrive/Colab_Notebooks/accident/accident_0000638.mp4,0.0
/content/drive/MyDrive/Colab_Notebooks/nonaccident/nonaccident_0002143.mp4,1.0
/content/drive/MyDrive/Colab_Notebooks/accident/accident_0000372.mp4,0.0
/content/drive/MyDrive/Colab_Notebooks/accident/accident_0000419.mp4,0.0
/content/drive/MyDrive/Colab_Notebooks/nonaccident/nonaccident_0001675.mp4,1.0
/content/drive/MyDrive/Colab_Notebooks/accident/accident_0000307.mp4,0.0


  


  Could someone tell me how do I modify the code to read the video clips instead of images ?

  


• How can I broadcast a Twilio (group room) video conference ?

  2 novembre 2022, par Danish

  I have been trying different solutions/approaches for last few days and posting here to get some more options.

  



  I am planning to use Twilio's Programmable Video solution with Group room (upto 50 participants) to allow hosts/presenters to join video conference meeting. Right now I am able to setup and get started with quickstart example here

  



  Also, in parallel, I need to be able to broadcast that live meeting room through RTMP/HLS (audio+video) streaming that can be viewed (through another portal) by large user set (ideally around 10k+)

  



  Twilio support team suggested - integrating Twilio Video with Wowza using screenscrapping and ffmpeg (i.e. render the video into a (maybe virtual) screen, capture it and send to Wowza using ffmpeg.)

  



  I am new to both ffmpeg and wowza. However, I am looking to see high level solution which can be used to achieve this (with minimum hanging parts).

  



  Ideal flow i can think of is

  



  

  • Start a Twilio video conference room with actual presenters

  


  • all participants share there video and audio tracks with each other in the room - standard behavior

  


  • a new ghost-participant joins-in from stand-alone server (from wowza or SIP) and subscribes to audio-video tracks in conference.

  


  • arrange & combine those tracks into one channel and stream it as HLS

  


  • this ghost-participant will also be able to identify dominant speaker and re-arrange video tracks to show current speaker occupying bigger space (just like zoom)

  


  • and of course, this streamed video (with audio) will be one-way and viewers cannot interact. Just see be able to see the room broadcasting.

  


  



  Also/FYI
I have tried exploring few more options below (with no success)

  



  


  1. a blog here explain how can we broadcast a twilio audio
conference. At the bottom it also mention about video streaming. but, I am not getting how can we use SIP to get video stream that can be broadcasted. Also, how can I manipulate video tracks to show dominant speakers in full screen view Vs other participant in thumbnail view i.e. custom layouts.

  


  2. I also explored option of Wowza WebRTC streaming to achieve this, but it has got other technical challenges with setting up video room with multiple participants. I am going to reach out to them to see if this use-case is possible with their offering.

  


  3. I am looking for something as simple as here - dial as guest participant into video room from standalone streaming server/wowza server. And get output as single stream to broadcast.

  


  



  Any other pointers or links to solutions would be helpful.

  


• Open Video Conference registration is now open !

  15 juillet 2010, par Basil Gohar — Announcements, conference, open video conference

  Registration for this year’s Open Video Conference (OVC) is now open ! Register on or before August 1st, 2010 to get the early bird discount. This year’s OVC will be held in New York City from October 1 – 2, 2010. … Read more »