No weekly build this week

Due to an error yesterday, we reverted back to an earlier model from about a week ago.  Quality of a build this week would be about the same as last week.

Classifiers

We are building and training some fine-grained classifiers using some proven and newer image classification networks (such as resnet, vgg19, efficientnet).  We are testing which ones produce the best results.  The classifiers will help us improve our training by identifying features like sex, hair/eye color, wardrobe, etc.  The classifiers will be able to do thousands of images per minute where humans can only do a few.

Experiments in quality and network size

We have conducted a few experiments to see if we can improve the quality of the images by tweaking parameters in the network.  So far, nothing really jumps out, but here are a few highlights from the "deep fashion" dataset.  We will be looking at entirely different architectures from the latest research over the next few months.

That's all for this weeks update.  We are looking forward to next week's build where hopefully all the background work we have done in image classification we have engaged in over the last few weeks will begin to show up in the generated images.  Have a great weekend!