brisk crane
Disclaimer: every bit of code used is generated by gpt with small modifications here and there from me if you dont like it then dont read the rest and then proceed get mad at me because im taking the easy route and using GPT to generate the code, its that simple. (putting this here due to other servers highly disliking my use of gpt to write code...)
Just a small and simple image to tag (like e926 tags) model using resnet50 and the default weights, planning on seeing if "efficientnet_b3" will give more accurate tag results but im also limited to 4GB of VRAM and can train 1 epoch with ~2k 512x512 images into this model in ~9 mins, though i need to get a better model handling system to handle models trained with different parameters without having to rename the model every time. i will show results soon after i train a sfw version of it :P