Face/people detection/tracking at scale like the Chinese are doing. Despite my moral/ethical compass revolting on this, it is one of the most impactful applications so far. It is not just face detection that it does, there are other more complicated things like gait/activity based classification. Route tracking, people interaction tracking, etc. while some of this is just hype, clearly a lot of money is flowing to solve these Chinese government problems.
Data Generation.
All kind of dvelopments arround GANs are really exciting.
I bet on GANs on videogames.soon for generating content the same way fractals did it 15/20 years ago (see The Sentinel, Archipelagos...)
When I first started my adventure into deep learning I became convinced that even a simple dense network could be used for storing and generating variations of 3D content.
I never took it beyond a simple dense network for a blade of grass, but I'd like to re-approach this using Tensorflow.JS. You can see the stupid simple POC at: https://jacob-ebey.github.io/gen-3d/ if you're interested.
Having experience working in CAD/CAM software, I have a feeling that the struggle in an approach like this will be maintaining meaningful output topology when adjusting parameters.
The simple approach in the grass POC was to have a single network input that was between 0 and 1 for each grass model. If we had 4 grass models, the training batch would have 4 entries with the first input being 0, the second being 0.25, etc... Then to generate a "grass" model, you could feed in any value between 0 and 1 to get a new model.
The catch here is that to maintain a meaningful output, the input models must have the same "shape". I.e the same number of vertices and indices as the indices from a base model are mapped to the new vertex values spat out by the network.
Text generation is neat. The first few iterations are nonsensical but then real words start appearing, then phrases start making sense, then the correct punctuation comes along and the machine is spitting out sentences, and before you know it the machine has created readable paragraphs. I have no idea what I'm doing, but it blows my mind.
Deepfakes. Could be used in TV/film for deceased actors or actors who have aged significantly from the original. E.g. Doctor Who won't have the budget of Blade Runner 2049 to resurrect previous Doctors using CGI.
I think there is going to be a revolution in using synthetic imagery / 3D renderings to train neural networks. Right now the biggest limiting factor imo is acquiring data sets and labeling them. That problem vanishes if you generate photorealistic images. Why this isn't a huge, huge, huge thing already I don't know, but I'm doing it.
8 comments
[ 3.6 ms ] story [ 24.2 ms ] threadML could be the human augmentation we need to Cure All The Things.
I never took it beyond a simple dense network for a blade of grass, but I'd like to re-approach this using Tensorflow.JS. You can see the stupid simple POC at: https://jacob-ebey.github.io/gen-3d/ if you're interested.
Having experience working in CAD/CAM software, I have a feeling that the struggle in an approach like this will be maintaining meaningful output topology when adjusting parameters.
The simple approach in the grass POC was to have a single network input that was between 0 and 1 for each grass model. If we had 4 grass models, the training batch would have 4 entries with the first input being 0, the second being 0.25, etc... Then to generate a "grass" model, you could feed in any value between 0 and 1 to get a new model.
The catch here is that to maintain a meaningful output, the input models must have the same "shape". I.e the same number of vertices and indices as the indices from a base model are mapped to the new vertex values spat out by the network.