0

I am looking for a video analysis tool. I would like to know can tensorflow object detection API tell me how many T-shirts appear in a movie? If not, what tool can do it?

Thanks

Renjie Zhang
  • 3
  • 2

1 Answers1

1

Videos are a just a sequence of images. You can use a number of libraries to extract the individual frames: OpenCV, FFMPEG (command line), and others. Note that Tensorflow itself doesn't provide an object detection API, I think what you mean to say is to use an Object Detection API that was built using Tensorflow. Most of those APIs will be expecting one, or a batch, of images and it will be up to you to extract the frames, run them through, and rebuild the video at the end.

David Parks
  • 30,789
  • 47
  • 185
  • 328
  • Here is the link of the object detected API: https://github.com/tensorflow/models/tree/master/research/object_detection – Renjie Zhang Apr 09 '18 at 19:30
  • Pretty standard stuff, trained on the Coco it looks like. Though I haven't used it I'm 99% sure you feed it images. Splitting them up is your responsibility, e.g.: https://stackoverflow.com/questions/33311153/python-extracting-and-saving-video-frames – David Parks Apr 09 '18 at 21:07
  • so your suggestion is to use Opencv.captureVideo to convert a video into many images and analysis each image? – Renjie Zhang Apr 10 '18 at 19:11
  • Personally, I find OpenCV too much of a pain to install for just reading frames from a video file, there are more simple python modules you can install with pip to do this, but irrespective of which library you use for video frame capture, yes, video analysis on a per-frame basis is the right approach. – David Parks Apr 10 '18 at 19:22