File size: 5,441 Bytes
ec08a0d
 
 
 
 
 
 
 
 
 
53f9d01
 
ec08a0d
e1d0b29
ec08a0d
125bdfb
ec08a0d
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
b80dc78
 
 
ec08a0d
 
 
a1593c2
 
 
 
 
 
 
 
 
 
 
ec08a0d
 
6b79de9
fe03815
 
 
0a01cc6
ec08a0d
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
d786888
 
ec08a0d
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
d786888
 
ec08a0d
 
 
 
 
 
 
 
 
 
 
6b79de9
 
 
 
 
 
 
 
d786888
ec08a0d
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
949e2ae
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
64
65
66
67
68
69
70
71
72
73
74
75
76
77
78
79
80
81
82
83
84
85
86
87
88
89
90
91
92
93
94
95
96
97
98
99
100
101
102
103
104
105
106
107
108
109
110
111
112
113
114
115
116
117
118
119
120
121
122
123
124
125
126
127
128
129
130
131
132
133
134
135
136
137
138
139
140
141
142
143
144
145
146
147
148
149
150
151
152
153
154
155
156
157
158
159
160
161
162
163
164
165
166
167
168
169
import gradio as gr
import numpy as np
import torch
from diffusers import StableDiffusionInpaintPipeline
from PIL import Image
from segment_anything import SamPredictor, sam_model_registry, SamAutomaticMaskGenerator
from diffusers import ControlNetModel
from diffusers import UniPCMultistepScheduler
from controlnet_inpaint import StableDiffusionControlNetInpaintPipeline
import colorsys

sam_checkpoint = "sam_vit_h_4b8939.pth"
model_type = "vit_h"
device = "cpu"


sam = sam_model_registry[model_type](checkpoint=sam_checkpoint)
sam.to(device=device)
predictor = SamPredictor(sam)
mask_generator = SamAutomaticMaskGenerator(sam)

# pipe = StableDiffusionInpaintPipeline.from_pretrained(
#     "stabilityai/stable-diffusion-2-inpainting",
#     torch_dtype=torch.float16,
# )
# pipe = pipe.to("cuda")

controlnet = ControlNetModel.from_pretrained(
    "lllyasviel/sd-controlnet-seg",
    torch_dtype=torch.float16,
)
pipe = StableDiffusionControlNetInpaintPipeline.from_pretrained(
    "runwayml/stable-diffusion-inpainting",
    controlnet=controlnet,
    torch_dtype=torch.float16,
)
pipe.scheduler = UniPCMultistepScheduler.from_config(pipe.scheduler.config)
#pipe.enable_model_cpu_offload()
#pipe.enable_xformers_memory_efficient_attention()
pipe = pipe.to(device)


with gr.Blocks() as demo:
    gr.Markdown("# StableSAM: Stable Diffusion + Segment Anything Model")
    gr.Markdown(
        """
    To try the demo, upload an image and select object(s) you want to inpaint.
    Write a prompt & a negative prompt to control the inpainting.
    Click on the "Submit" button to inpaint the selected object(s).
    Check "Background" to inpaint the background instead of the selected object(s).

    If the demo is slow, clone the space to your own HF account and run on a GPU.
    """
    )
    selected_pixels = gr.State([])
    with gr.Row():
        input_img = gr.Image(label="Input")
        mask_img = gr.Image(label="Mask", interactive=False)
        seg_img = gr.Image(label="Segmentation", interactive=False)
        output_img = gr.Image(label="Output", interactive=False)

    with gr.Row():
        prompt_text = gr.Textbox(lines=1, label="Prompt")
        negative_prompt_text = gr.Textbox(lines=1, label="Negative Prompt")
        is_background = gr.Checkbox(label="Background")

    with gr.Row():
        submit = gr.Button("Submit")
        clear = gr.Button("Clear")

    def generate_mask(image, bg, sel_pix, evt: gr.SelectData):
        sel_pix.append(evt.index)
        predictor.set_image(image)
        input_point = np.array(sel_pix)
        input_label = np.ones(input_point.shape[0])
        mask, _, _ = predictor.predict(
            point_coords=input_point,
            point_labels=input_label,
            multimask_output=False,
        )
        # clear torch cache
        torch.cuda.empty_cache()
        if bg:
            mask = np.logical_not(mask)
        mask = Image.fromarray(mask[0, :, :])
        segs = mask_generator.generate(image)
        boolean_masks = [s["segmentation"] for s in segs]
        finseg = np.zeros((boolean_masks[0].shape[0], boolean_masks[0].shape[1], 3), dtype=np.uint8)
        # Loop over the boolean masks and assign a unique color to each class
        for class_id, boolean_mask in enumerate(boolean_masks):
            hue = class_id * 1.0 / len(boolean_masks)
            rgb = tuple(int(i * 255) for i in colorsys.hsv_to_rgb(hue, 1, 1))
            rgb_mask = np.zeros((boolean_mask.shape[0], boolean_mask.shape[1], 3), dtype=np.uint8)
            rgb_mask[:, :, 0] = boolean_mask * rgb[0]
            rgb_mask[:, :, 1] = boolean_mask * rgb[1]
            rgb_mask[:, :, 2] = boolean_mask * rgb[2]
            finseg += rgb_mask

        torch.cuda.empty_cache()

        return mask, finseg

    def inpaint(image, mask, seg_img, prompt, negative_prompt):
        image = Image.fromarray(image)
        mask = Image.fromarray(mask)
        seg_img = Image.fromarray(seg_img)

        image = image.resize((512, 512))
        mask = mask.resize((512, 512))
        seg_img = seg_img.resize((512, 512))

        output = pipe(
            prompt,
            image,
            mask,
            seg_img,
            negative_prompt=negative_prompt,
            num_inference_steps=20,
        ).images[0]
        torch.cuda.empty_cache()
        return output

    def _clear(sel_pix, img, mask, seg, out, prompt, neg_prompt, bg):
        sel_pix = []
        img = None
        mask = None
        seg = None
        out = None
        prompt = ""
        neg_prompt = ""
        bg = False
        return img, mask, seg, out, prompt, neg_prompt, bg

    input_img.select(
        generate_mask,
        [input_img, is_background, selected_pixels],
        [mask_img, seg_img],
    )
    submit.click(
        inpaint,
        inputs=[input_img, mask_img, seg_img, prompt_text, negative_prompt_text],
        outputs=[output_img],
    )
    clear.click(
        _clear,
        inputs=[
            selected_pixels,
            input_img,
            mask_img,
            seg_img,
            output_img,
            prompt_text,
            negative_prompt_text,
            is_background,
        ],
        outputs=[
            input_img,
            mask_img,
            seg_img,
            output_img,
            prompt_text,
            negative_prompt_text,
            is_background,
        ],
    )

if __name__ == "__main__":
    demo.launch()