Offline, fast image recognition? Where do I get started?

Accelerated Computing Intelligent Video Analytics DeepStream SDK
1

I have a library of 10 million photos from sporting events that we cover and I want to train an AI when to trigger (take the photo) based on the content of video coming in from a webcam. This is sport specific. I’m an experienced web developer, primarily JS and PHP.

The solution must be offline because the venues are often in locations with low or slow internet.

Based on my research, it would be best if I had unsupervised training (feed all of the images in) and then supervised training (show it images that are positive (take the shot), and images that are negative (don’t take the shot).

Then, I need some hardware and a laptop to read the input and trigger the API calls to the camera to take the shot.

So:

  1. Train a model using the 10 million photos (I’m not sure what to do there, there’s so many options, but I need it to be offline.
  2. Purchase hardware for on-site processing. (What hardware do I need? Will a laptop work?)
  3. Build the API integration with Sony’s camera AI (I can handle this already no problem).

So for 1 + 2, what’s the best way for me to proceed?

3

Are you asking for a proposal of how to train your model with your images? For different models, the model projects also provide training process in the model projects, it is a PyTorch script for most models. You may consult the author of the model for the training methods.

The hardware needed for the model training depends on your AI model and your requirement for the performance. It should be case by case.

4

There is no update from you for a period, assuming this is not an issue anymore. Hence we are closing this topic. If need further support, please open a new one. Thanks.

5

This topic was automatically closed 14 days after the last reply. New replies are no longer allowed.