0

I have a 10 second video (60fps) of a heart taken using m-mode ultrasound and I was hoping to train a prediction model on these with each video tagged with three custom labels. I am not interested in doing a classification on each individual frame as I could do that myself, rather classify the 10 second video based on each frame and changes (eg movement) between the series of ~600 frames.

Does either Clarifai Video V2 or Google's Video Intelligence offer training/prediction like this with custom labels?

Byron Rogers
  • 63
  • 5

1 Answers1

0

You can definitely predict using (any amount of) custom labels using Clarifai. Create a new model, add your own three labels (called concepts) to it and train it.

Regarding how video prediction is done:

The Video API analyzes video and predicts what's inside of it.

The Video API will return a list of predicted concepts for every frame of video. Video is processed at 1 frame per second. This means you will receive a list of concepts for every second of your video.

Please see more here.

Community
  • 1
  • 1
Rok Povsic
  • 4,626
  • 5
  • 37
  • 53
  • I saw that. Thanks. The issue I have is two fold 1) They are only using one frame per second. If we are taking our video at 60fps, 59 of these are unused and they are only using 10 frames in a 10 second video. That seems like a waste of potential information for me. 2) The prediction can only be performed on pre-trained models for video. You cant create a custom model with custom data. – Byron Rogers Aug 25 '17 at 09:23