Neural-Net-Based Tracking

< Previous | Contents | Manuals Home | Boris FX | Next >

‌Neural-Net-Based Tracking ‌

Many recent technical advances in image processing have been based on a technology called neural nets, which is roughly based on elements of the behavior of neurons in brains. These advances include finding faces to focus on in cameras, identifying specific people in pictures, reading postal codes, sorting out pictures of cats versus dogs, up to self-driving cars(!).

Important : The pixel aspect ratio should be 1.0 for neural-net processing. Use the image preprocessor’s Output tab to set up resampling to make that the case.

TL;DR : The neural nets are very slow for auto-tracking! Better for supervised. If you have a suitable NVIDIA GPU, you may be able to increase performance by installing the Tensorflow-GPU overlay, see the GPUs and Performance section. If neural-net tracking helps you on a specific task (checkerboards or X tracking marks?), great! If not, stick with the usual stuff. It’s just another tool in the bin.

SynthEyes has neural-net-based tracking as well, for both automatic and supervised tracking. While the hype may lead us all to believe that anything can be accomplished, the reality is more complex.

Why It’s Not Just Cats and Dogs Currently Distributed Neural Nets X-Finder Checkerboard Automatic Tracking Controls Supervised Tracking Controls GPUs and Performance Installing the Tensorflow-for-GPUs Overlay Preferences Auto Tracking Preferences Neural Supervised Preferences GPU Preferences Suggested Usage

< Previous | Contents | Manuals Home | Boris FX | Next >