Convolutional Neural Network in Substance 3D Designer
The graph in action.
Did you know that there is a difference between a "Eskimo dog" and a "Husky"?
The Top - Level graph of the network, it's too large to get uploaded to Artstation, so I had to get a bit creative and rendered a flythrough.
Convolutional Neural Network in Substance 3D Designer
Hi everyone,
I hope you are well and healthy. During nodevember, I asked myself if it is possible to implement a fully functional convolutional neural network in Substance 3D Designer?
The answer is YES! I implemented the famous VGG16 Model from scratch using the pretrained weights from the original model trained on the #ImageNet dataset predicting 1000 different classes.
The result is probably the largest graph I've ever done, here are some impressive statistics about the node:
- 8488 Pixelprocessors
- 4225 Valueprocessors
- 138357544 parameters
- .sbsar size 594.65MB
- computetime 3.5 seconds
You'll find a breakdown of the graph on my artstation blog (https://www.artstation.com/blogs/marcovitale/Xnn00/convolutional-neural-network-in-substance-3d-designer)
Since this was a personal challenge the graph needs a lot of cleanup before I can share it publicly, so please have some patience but I will upload the complete framework of nodes created for CNNs soon.
Until then I uploaded the .sbsar, the high resolution image of the graph and a list of available classes to the GoogleDrive (https://tinyurl.com/bdeptnkx) feel free to download it and play with it as much as you want.
Stay healthy and creative Marco