Create ref_in_gp3.py

tasks/ref_in_gp3.py

@@ -0,0 +1,108 @@
+# --------------------------------------------------------
+# X-Decoder -- Generalized Decoding for Pixel, Image, and Language
+# Copyright (c) 2022 Microsoft
+# Licensed under The MIT License [see LICENSE for details]
+# Written by Jianwei Yang (jianwyan@microsoft.com)
+# --------------------------------------------------------
+import os
+import openai
+import torch
+import numpy as np
+from scipy import ndimage
+from PIL import Image
+from utils.inpainting import pad_image, crop_image
+from torchvision import transforms
+from utils.visualizer import Visualizer
+from diffusers import StableDiffusionInpaintPipeline
+from detectron2.utils.colormap import random_color
+from detectron2.data import MetadataCatalog
+
+
+t = []
+t.append(transforms.Resize(512, interpolation=Image.BICUBIC))
+transform = transforms.Compose(t)
+metadata = MetadataCatalog.get('ade20k_panoptic_train')
+
+pipe = StableDiffusionInpaintPipeline.from_pretrained(
+    # "stabilityai/stable-diffusion-2-inpainting",
+    "runwayml/stable-diffusion-inpainting",
+    revision="fp16", 
+    torch_dtype=torch.float16,
+).to("cuda")
+
+prompts = []
+prompts.append("instruction: remove the person, task: (referring editing), source: [person], target:<clean and empty scene>.")
+prompts.append("instruction: remove the person in the middle, task: (referring editing), source: [person in the middle], target:<clean and empty scene>.")
+prompts.append("instruction: remove the dog on the left side, task: (referring editing), source: [dog on the left side], target:<clean and empty scene>.")
+prompts.append("instruction: change the apple to a pear, task: (referring editing), source: [apple], target: <pear>.")
+prompts.append("instruction: change the red apple to a green one, task: (referring editing), source: [red apple], target: <green apple>.")
+prompts.append("instruction: change the color of bird's feathers from white to blue, task: (referring editing), source: [white bird], target: <blue bird>.")
+prompts.append("instruction: replace the dog with a cat, task: (referring editing), source: [dot], target: <cat>.")
+prompts.append("instruction: replace the red apple with a green one, task: (referring editing), source: [red apple], target: <green apple>.")
+
+openai.api_type = "azure"
+openai.api_base = "https://xdecoder.openai.azure.com/"
+openai.api_version = "2022-12-01"
+openai.api_key = os.environ["OPENAI_API_KEY"]
+
+def get_gpt3_response(prompt):
+    response = openai.Completion.create(
+      engine="text001",
+      prompt=prompt,
+      temperature=0.7,
+      max_tokens=512,
+      top_p=1,
+      frequency_penalty=0,
+      presence_penalty=0,
+    )
+    
+    return response
+
+def referring_inpainting_gpt3(model, image, instruction, *args, **kwargs):    
+    # convert instruction to source and target
+    instruction = instruction.replace('.', '')
+    print(instruction)
+    resp = get_gpt3_response(' '.join(prompts) + ' instruction: ' + instruction + ',')
+    resp_text = resp['choices'][0]['text']
+    print(resp_text)
+    ref_text = resp_text[resp_text.find('[')+1:resp_text.find(']')]
+    inp_text = resp_text[resp_text.find('<')+1:resp_text.find('>')]
+
+    model.model.metadata = metadata
+    texts = [[ref_text if ref_text.strip().endswith('.') else (ref_text.strip() + '.')]]
+    image_ori = crop_image(transform(image))
+
+    with torch.no_grad():
+        width = image_ori.size[0]
+        height = image_ori.size[1]
+        image = np.asarray(image_ori)
+        image_ori_np = np.asarray(image_ori)
+        images = torch.from_numpy(image.copy()).permute(2,0,1).cuda()
+
+        batch_inputs = [{'image': images, 'height': height, 'width': width, 'groundings': {'texts': texts}}]        
+        outputs = model.model.evaluate_grounding(batch_inputs, None)
+        visual = Visualizer(image_ori_np, metadata=metadata)
+
+        grd_mask = (outputs[0]['grounding_mask'] > 0).float().cpu().numpy()
+        for idx, mask in enumerate(grd_mask):
+            color = random_color(rgb=True, maximum=1).astype(np.int32).tolist()
+            demo = visual.draw_binary_mask(mask, color=color, text=texts[idx])
+        res = demo.get_image()
+    
+    if inp_text not in ['no', '']:
+        image_crop = image_ori
+        struct2 = ndimage.generate_binary_structure(2, 2)
+        mask_dilated = ndimage.binary_dilation(grd_mask[0], structure=struct2, iterations=3).astype(grd_mask[0].dtype)
+        mask = Image.fromarray(mask_dilated * 255).convert('RGB')
+        image_and_mask = {
+            "image": image_crop,
+            "mask": mask,
+        }
+        # images_inpainting = inpainting(inpainting_model, image_and_mask, inp_text, ddim_steps, num_samples, scale, seed)
+        width = image_ori.size[0]; height = image_ori.size[1]
+        images_inpainting = pipe(prompt = inp_text.strip(), image=image_and_mask['image'], mask_image=image_and_mask['mask'], height=height, width=width).images
+        torch.cuda.empty_cache()
+        return images_inpainting[0]
+    else:
+        torch.cuda.empty_cache()
+        return Image.fromarray(res)