[Feature Request] Generate classification dataset automatically about sd_dreambooth_extension

[YakuzaSuske]

Shiv's repo does this where it makes a set of 50 images of the classification prompt if the class folder is empty. Right now in the webui I have to generate the prompt manually in txt2img and then move them to a new folder that I feed into the extension. Would be good if that process can be automated.

So lets say i want to train an anime girl, do i just search up anime girls and download them and set those as my class images? or do i have to do something else? i think thats why im getting bad results, because i don't have class images.

[d8ahazard]

Shiv's repo does this where it makes a set of 50 images of the classification prompt if the class folder is empty. Right now in the webui I have to generate the prompt manually in txt2img and then move them to a new folder that I feed into the extension. Would be good if that process can be automated.

This does this as well, you just specify the number of class images and prompt, and it will generate the images.

[ShinkoNet]

Shiv's repo does this where it makes a set of 50 images of the classification prompt if the class folder is empty. Right now in the webui I have to generate the prompt manually in txt2img and then move them to a new folder that I feed into the extension. Would be good if that process can be automated.

This does this as well, you just specify the number of class images and prompt, and it will generate the images.

"Total number of classification images to use. Set to 0 to disable." right?
Oops. You can close this then. Thanks for the info

[YakuzaSuske]

when training with classification images, i get CUDA out of memory.

CUDA SETUP: Loading binary C:\SUPER_SD_2.0\stable-diffusion-webui-master\venv\lib\site-packages\bitsandbytes\libbitsandbytes_cuda116.dll...
The config attributes {'set_alpha_to_one': False, 'skip_prk_steps': True, 'steps_offset': 1} were passed to DDPMScheduler, but are not expected and will be ignored. Please verify your scheduler_config.json configuration file.
Scheduler Loaded
Allocated: 0.2GB
Reserved: 0.2GB

Total target lifetime optimization steps = 1000
CPU: False Adam: True, Prec: fp16, Prior: True, Grad: True, TextTr: True
Allocated: 3.8GB
Reserved: 3.9GB

Steps: 0%| | 0/1000 [00:00<?, ?it/s]Error completing request
Arguments: ('Sakie', 'C:\Users\Hector\Pictures\New folder\Sakie\New folder (2)', 'C:\SUPER_SD_2.0\stable-diffusion-webui-master\models\dreambooth\Sakie\New folder', 'Sakie', 'person', '', '', 1.0, 7.5, 40.0, 57, 512, False, True, 1, 1, 1, 1000, 1, True, 5e-06, False, 'constant', 0, True, 0.9, 0.999, 0.01, 1e-08, 1, 100, 500, 'fp16', True, '', False) {}
Traceback (most recent call last):
File "C:\SUPER_SD_2.0\stable-diffusion-webui-master\modules\ui.py", line 185, in f
res = list(func(*args, **kwargs))
File "C:\SUPER_SD_2.0\stable-diffusion-webui-master\webui.py", line 54, in f
res = func(*args, **kwargs)
File "C:\SUPER_SD_2.0\stable-diffusion-webui-master\extensions\sd_dreambooth_extension\dreambooth\dreambooth.py", line 256, in start_training
trained_steps = main(config)
File "C:\SUPER_SD_2.0\stable-diffusion-webui-master\extensions\sd_dreambooth_extension\dreambooth\train_dreambooth.py", line 766, in main
accelerator.backward(loss)
File "C:\SUPER_SD_2.0\stable-diffusion-webui-master\venv\lib\site-packages\accelerate\accelerator.py", line 882, in backward
self.scaler.scale(loss).backward(**kwargs)
File "C:\SUPER_SD_2.0\stable-diffusion-webui-master\venv\lib\site-packages\torch_tensor.py", line 396, in backward
torch.autograd.backward(self, gradient, retain_graph, create_graph, inputs=inputs)
File "C:\SUPER_SD_2.0\stable-diffusion-webui-master\venv\lib\site-packages\torch\autograd_init_.py", line 173, in backward
Variable.execution_engine.run_backward( # Calls into the C++ engine to run the backward pass
File "C:\SUPER_SD_2.0\stable-diffusion-webui-master\venv\lib\site-packages\torch\autograd\function.py", line 253, in apply
return user_fn(self, *args)
File "C:\SUPER_SD_2.0\stable-diffusion-webui-master\venv\lib\site-packages\torch\utils\checkpoint.py", line 146, in backward
torch.autograd.backward(outputs_with_grad, args_with_grad)
File "C:\SUPER_SD_2.0\stable-diffusion-webui-master\venv\lib\site-packages\torch\autograd_init.py", line 173, in backward
Variable._execution_engine.run_backward( # Calls into the C++ engine to run the backward pass
RuntimeError: CUDA out of memory. Tried to allocate 1024.00 MiB (GPU 0; 12.00 GiB total capacity; 9.67 GiB already allocated; 0 bytes free; 10.37 GiB reserved in total by PyTorch) If reserved memory is >> allocated memory try setting max_split_size_mb to avoid fragmentation. See documentation for Memory Management and PYTORCH_CUDA_ALLOC_CONF

Steps: 0%| | 0/1000 [00:09<?, ?it/s]