#credits to Valhalla for his space: https://huggingface.co/spaces/valhalla/glide-text2im that was used to build this app. #credits to the researchers of OpenAI for providing the text2img algorithm import os os.system('pip install -e .') import gradio as gr import base64 from io import BytesIO # from fastapi import FastAPI from PIL import Image import torch as th from glide_text2im.download import load_checkpoint from glide_text2im.model_creation import ( create_model_and_diffusion, model_and_diffusion_defaults, model_and_diffusion_defaults_upsampler ) # print("Loading models...") # app = FastAPI() # This notebook supports both CPU and GPU. # On CPU, generating one sample may take on the order of 20 minutes. # On a GPU, it should be under a minute. has_cuda = th.cuda.is_available() device = th.device('cpu' if not has_cuda else 'cuda') # # Create base model. # options = model_and_diffusion_defaults() # options['use_fp16'] = has_cuda # options['timestep_respacing'] = '40' # use 100 diffusion steps for fast sampling (Previous it was 100) # model, diffusion = create_model_and_diffusion(**options) # model.eval() # if has_cuda: # model.convert_to_fp16() # model.to(device) # # model.load_state_dict(load_checkpoint('base', device)) # model.load_state_dict(th.load("base.pt", map_location=device)) # print('total base parameters', sum(x.numel() for x in model.parameters())) # # Create upsampler model. # options_up = model_and_diffusion_defaults_upsampler() # options_up['use_fp16'] = has_cuda # options_up['timestep_respacing'] = 'fast27' # use 27 diffusion steps for very fast sampling # model_up, diffusion_up = create_model_and_diffusion(**options_up) # model_up.eval() # if has_cuda: # model_up.convert_to_fp16() # model_up.to(device) # # model_up.load_state_dict(load_checkpoint('upsample', device)) # model.load_state_dict(th.load("upsample.pt", map_location=device)) # print('total upsampler parameters', sum(x.numel() for x in model_up.parameters())) base_timestep_respacing = '100' #@param {type:"string"} sr_timestep_respacing = 'fast27' #@title Create base model. glide_path = 'glide-ft-13x625.pt' #@param {type:"string"} import os options = model_and_diffusion_defaults() options['use_fp16'] = has_cuda options['timestep_respacing'] = base_timestep_respacing # use 100 diffusion steps for fast sampling model, diffusion = create_model_and_diffusion(**options) if len(glide_path) > 0: assert os.path.exists( glide_path ), f"Failed to resume from {glide_path}, file does not exist." weights = th.load(glide_path, map_location="cpu") model, diffusion = create_model_and_diffusion(**options) model.load_state_dict(weights) print(f"Resumed from {glide_path} successfully.") else: model, diffusion = create_model_and_diffusion(**options) model.load_state_dict(load_checkpoint("base", device)) model.eval() if has_cuda: model.convert_to_fp16() model.to(device) print('total base parameters', sum(x.numel() for x in model.parameters())) #@title Create upsampler model. sr_glide_path = "glide-ft-16x5000.pt" #@param {type:"string"} options_up = model_and_diffusion_defaults_upsampler() options_up['use_fp16'] = has_cuda options_up['timestep_respacing'] = sr_timestep_respacing # use 27 diffusion steps for very fast sampling if len(sr_glide_path) > 0: assert os.path.exists( sr_glide_path ), f"Failed to resume from {sr_glide_path}, file does not exist." weights = th.load(sr_glide_path, map_location="cpu") model_up, diffusion_up = create_model_and_diffusion(**options_up) model_up.load_state_dict(weights) print(f"Resumed from {sr_glide_path} successfully.") else: model_up, diffusion_up = create_model_and_diffusion(**options) model_up.load_state_dict(load_checkpoint("upsample", device)) if has_cuda: model_up.convert_to_fp16() model_up.to(device) print('total upsampler parameters', sum(x.numel() for x in model_up.parameters())) def get_images(batch: th.Tensor): """ Display a batch of images inline. """ scaled = ((batch + 1)*127.5).round().clamp(0,255).to(th.uint8).cpu() reshaped = scaled.permute(2, 0, 3, 1).reshape([batch.shape[2], -1, 3]) return Image.fromarray(reshaped.numpy()) # Create a classifier-free guidance sampling function guidance_scale = 8.0 def model_fn(x_t, ts, **kwargs): half = x_t[: len(x_t) // 2] combined = th.cat([half, half], dim=0) model_out = model(combined, ts, **kwargs) eps, rest = model_out[:, :3], model_out[:, 3:] cond_eps, uncond_eps = th.split(eps, len(eps) // 2, dim=0) half_eps = uncond_eps + guidance_scale * (cond_eps - uncond_eps) eps = th.cat([half_eps, half_eps], dim=0) return th.cat([eps, rest], dim=1) def to_base64(pil_image): buffered = BytesIO() pil_image.save(buffered, format="JPEG") return base64.b64encode(buffered.getvalue()) # @app.get("/") def read_root(): return {"glide!"} # @app.get("/{generate}") def sample(prompt): # Sampling parameters batch_size = 1 # Tune this parameter to control the sharpness of 256x256 images. # A value of 1.0 is sharper, but sometimes results in grainy artifacts. upsample_temp = 1.0 ############################## # Sample from the base model # ############################## # Create the text tokens to feed to the model. tokens = model.tokenizer.encode(prompt) tokens, mask = model.tokenizer.padded_tokens_and_mask( tokens, options['text_ctx'] ) # Create the classifier-free guidance tokens (empty) full_batch_size = batch_size * 2 uncond_tokens, uncond_mask = model.tokenizer.padded_tokens_and_mask( [], options['text_ctx'] ) # Pack the tokens together into model kwargs. model_kwargs = dict( tokens=th.tensor( [tokens] * batch_size + [uncond_tokens] * batch_size, device=device ), mask=th.tensor( [mask] * batch_size + [uncond_mask] * batch_size, dtype=th.bool, device=device, ), ) # Sample from the base model. model.del_cache() samples = diffusion.p_sample_loop( model_fn, (full_batch_size, 3, options["image_size"], options["image_size"]), device=device, clip_denoised=True, progress=True, model_kwargs=model_kwargs, cond_fn=None, )[:batch_size] model.del_cache() ############################## # Upsample the 64x64 samples # ############################## tokens = model_up.tokenizer.encode(prompt) tokens, mask = model_up.tokenizer.padded_tokens_and_mask( tokens, options_up['text_ctx'] ) # Create the model conditioning dict. model_kwargs = dict( # Low-res image to upsample. low_res=((samples+1)*127.5).round()/127.5 - 1, # Text tokens tokens=th.tensor( [tokens] * batch_size, device=device ), mask=th.tensor( [mask] * batch_size, dtype=th.bool, device=device, ), ) # Sample from the base model. model_up.del_cache() up_shape = (batch_size, 3, options_up["image_size"], options_up["image_size"]) up_samples = diffusion_up.ddim_sample_loop( model_up, up_shape, noise=th.randn(up_shape, device=device) * upsample_temp, device=device, clip_denoised=True, progress=True, model_kwargs=model_kwargs, cond_fn=None, )[:batch_size] model_up.del_cache() image = get_images(up_samples) # image = to_base64(image) # return {"image": image} resized_image = image.resize((128, 128), Image.LANCZOS) return resized_image title = "Interactive demo: glide-text2im dermoscopic image generator" description = "Demo for the Finetuned version of OpenAI's GLIDE: Towards Photorealistic Image Generation and Editing with Text-Guided Diffusion Models. Please be aware that generation of the image will take up to 20 minutes, as CPU is used for the generation, and the resolution of generated images was limited for faster processing. For research purposes, we recommend using the finetuned model and weights provided here: https://github.com/Freiburg-AI-Research on your local GPU. Please cite our research paper with the title -Finetuning of GLIDE stable diffusion model for AI-based text-conditional image synthesis of dermoscopic images- when using the generator for your research." article = "

GLIDE: Towards Photorealistic Image Generation and Editing with Text-Guided Diffusion Models | Official Repo

" examples =["melanoma"] iface = gr.Interface(fn=sample, inputs=gr.inputs.Textbox(label='Which dermoscopic entity would you like to see? Choose one of the following one: "melanoma", "melanocytic nevi", "Actinic keratoses and intraepithelial carcinoma / Bowen disease, "benign keratosis-like lesions (solar lentigines / seborrheic keratoses and lichen-planus like keratoses", "basal cell carcinoma", "dermatofibroma", "vascular lesions"'), outputs=gr.outputs.Image(type="pil", label="Synthetic").style(height=128, width=128), title=title, description=description, article=article, examples=examples, enable_queue=True) iface.launch(debug=True)