Applying still image model to video

[Tom Tolleson]

Hello!

I've completed training on a model based on 11 sets of images of different fish species. VisionML is able to test fairly accurately on these.

However, my collaborators would like to apply this same object recognition to video files and live streaming video.

Is this possible from a model trained on still images? How would I take the next steps for that?

Regards,

Tom

[Jun (Cloud Platform Support)]

Hi Tom, 

As per features overview at [1][2], it looks that the custom trained image models can only be used for detecting individual objects in a given image. 

However, you can consider using the Cloud Video Intelligence API [3][4] for tracking objects detected in an input video. 

[1] https://cloud.google.com/vision/automl/object-detection/docs

[2] https://cloud.google.com/vertex-ai/docs/tutorials/image-recognition-custom/serving

[3] https://cloud.google.com/video-intelligence/docs/object-tracking

[4] https://cloud.google.com/vertex-ai/docs/tutorials/video-classification-automl/dataset