File size: 13,294 Bytes
5b6e775
5ade3f9
 
 
 
 
 
 
 
 
 
 
 
5b51ca6
5ade3f9
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
5b51ca6
 
5ade3f9
 
 
 
 
 
5b51ca6
5ade3f9
 
 
 
 
5b51ca6
5ade3f9
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
5b51ca6
 
 
 
 
 
5ade3f9
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
5b51ca6
5ade3f9
 
5b51ca6
5ade3f9
5b51ca6
5ade3f9
5b51ca6
 
 
5ade3f9
 
 
5b51ca6
5ade3f9
5b51ca6
5ade3f9
 
 
 
 
5b51ca6
 
 
 
 
 
 
5ade3f9
5b51ca6
 
 
 
 
 
 
 
5ade3f9
5b51ca6
 
 
 
5ade3f9
 
 
 
5b51ca6
 
 
 
 
 
 
5ade3f9
 
5b51ca6
 
 
 
 
 
 
 
 
5ade3f9
 
5b51ca6
 
 
 
 
 
5ade3f9
 
5b51ca6
 
 
 
 
5ade3f9
 
 
 
 
 
 
 
 
 
 
 
5b51ca6
 
 
 
5ade3f9
 
5b51ca6
 
 
 
 
 
 
 
 
 
 
 
5ade3f9
 
 
5b51ca6
5ade3f9
 
5b51ca6
5ade3f9
 
 
 
 
 
5b6e775
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
5ade3f9
5b6e775
 
 
5ade3f9
5b6e775
 
 
 
 
 
 
 
 
5ade3f9
5b6e775
 
 
5ade3f9
5b6e775
 
5ade3f9
5b6e775
5ade3f9
5b6e775
 
 
5b51ca6
5b6e775
 
5b51ca6
5b6e775
 
 
 
 
 
 
5b51ca6
5b6e775
 
 
5ade3f9
5b6e775
 
 
 
 
5b51ca6
5b6e775
 
5ade3f9
5b6e775
5ade3f9
 
5b6e775
5ade3f9
5b51ca6
 
 
 
 
 
 
 
 
 
 
 
5ade3f9
 
5b51ca6
 
 
 
 
 
 
 
 
 
5ade3f9
 
 
5b51ca6
 
 
0150383
5ade3f9
 
 
 
5b51ca6
 
 
 
 
5ade3f9
 
 
5b51ca6
 
 
 
 
 
 
 
 
5ade3f9
5b51ca6
 
 
 
 
 
 
 
 
5ade3f9
 
 
07775f7
5ade3f9
 
 
 
 
 
 
 
5b51ca6
5ade3f9
5b51ca6
5ade3f9
5b51ca6
 
 
 
 
 
5ade3f9
 
 
 
5b51ca6
 
 
5ade3f9
 
5b51ca6
 
 
 
 
 
 
 
 
 
 
 
 
 
5ade3f9
5b51ca6
5ade3f9
 
 
 
5b51ca6
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
170
171
172
173
174
175
176
177
178
179
180
181
182
183
184
185
186
187
188
189
190
191
192
193
194
195
196
197
198
199
200
201
202
203
204
205
206
207
208
209
210
211
212
213
214
215
216
217
218
219
220
221
222
223
224
225
226
227
228
229
230
231
232
233
234
235
236
237
238
239
240
241
242
243
244
245
246
247
248
249
250
251
252
253
254
255
256
257
258
259
260
261
262
263
264
265
266
267
268
269
270
271
272
273
274
275
276
277
278
279
280
281
282
283
284
285
286
287
288
289
290
291
292
293
294
295
296
297
298
299
300
301
302
303
304
305
306
307
308
309
310
311
312
313
314
315
316
317
318
319
320
321
322
323
324
325
326
327
328
329
330
331
332
333
334
335
336
337
338
339
340
341
342
343
344
345
346
347
348
349
350
351
352
353
354
355
356
357
358
359
360
361
362
363
364
365
366
367
368
369
370
371
372
373
374
375
376
377
378
379
380
381
382
383
384
385
386
387
388
389
390
391
392
393
394
395
396
397
398
399
400
401
402
403
404
405
406
407
408
409
410
411
412
413
414
415
import spaces
import gradio as gr
import random
import numpy as np
import os
import requests
import torch
import torchvision.transforms as T
from PIL import Image
from transformers import AutoProcessor, AutoModelForVision2Seq
import cv2
import ast


colors = [
    (0, 255, 0),
    (0, 0, 255),
    (255, 255, 0),
    (255, 0, 255),
    (0, 255, 255),
    (114, 128, 250),
    (0, 165, 255),
    (0, 128, 0),
    (144, 238, 144),
    (238, 238, 175),
    (255, 191, 0),
    (0, 128, 0),
    (226, 43, 138),
    (255, 0, 255),
    (0, 215, 255),
    (255, 0, 0),
]

color_map = {
    f"{color_id}": f"#{hex(color[2])[2:].zfill(2)}{hex(color[1])[2:].zfill(2)}{hex(color[0])[2:].zfill(2)}"
    for color_id, color in enumerate(colors)
}


def is_overlapping(rect1, rect2):
    x1, y1, x2, y2 = rect1
    x3, y3, x4, y4 = rect2

    return not (x2 < x3 or x1 > x4 or y2 < y3 or y1 > y4)


def draw_entity_boxes_on_image(image, entities, show=False, save_path=None, entity_index=-1):
    """_summary_

    Args:
        image (_type_): image or image path
        collect_entity_location (_type_): _description_
    """
    if isinstance(image, Image.Image):
        image_h = image.height
        image_w = image.width
        image = np.array(image)[:, :, [2, 1, 0]]
    elif isinstance(image, str):
        if os.path.exists(image):
            pil_img = Image.open(image).convert("RGB")
            image = np.array(pil_img)[:, :, [2, 1, 0]]
            image_h = pil_img.height
            image_w = pil_img.width
        else:
            raise ValueError(f"invaild image path, {image}")
    elif isinstance(image, torch.Tensor):
        image_tensor = image.cpu()
        reverse_norm_mean = torch.tensor([0.48145466, 0.4578275, 0.40821073])[
            :, None, None
        ]
        reverse_norm_std = torch.tensor([0.26862954, 0.26130258, 0.27577711])[
            :, None, None
        ]
        image_tensor = image_tensor * reverse_norm_std + reverse_norm_mean
        pil_img = T.ToPILImage()(image_tensor)
        image_h = pil_img.height
        image_w = pil_img.width
        image = np.array(pil_img)[:, :, [2, 1, 0]]
    else:
        raise ValueError(f"invaild image format, {type(image)} for {image}")

    if len(entities) == 0:
        return image

    indices = list(range(len(entities)))
    if entity_index >= 0:
        indices = [entity_index]

    # Not to show too many bboxes
    entities = entities[: len(color_map)]
    new_image = image.copy()
    previous_bboxes = []

    text_size = 1
    text_line = 1
    box_line = 3
    (c_width, text_height), _ = cv2.getTextSize(
        "F", cv2.FONT_HERSHEY_COMPLEX, text_size, text_line
    )
    base_height = int(text_height * 0.675)
    text_offset_original = text_height - base_height
    text_spaces = 3
    used_colors = colors
    color_id = -1

    for entity_idx, (entity_name, (start, end), bboxes) in enumerate(entities):
        color_id += 1
        if entity_idx not in indices:
            continue

        for bbox_id, (x1_norm, y1_norm, x2_norm, y2_norm) in enumerate(bboxes):
            orig_x1, orig_y1, orig_x2, orig_y2 = (
                int(x1_norm * image_w),
                int(y1_norm * image_h),
                int(x2_norm * image_w),
                int(y2_norm * image_h),
            )

            color = used_colors[color_id]
            new_image = cv2.rectangle(
                new_image,
                (orig_x1, orig_y1),
                (orig_x2, orig_y2),
                color,
                box_line,
            )

            l_o, r_o = (
                box_line // 2 + box_line % 2,
                box_line // 2 + box_line % 2 + 1,
            )
            x1 = orig_x1 - l_o
            y1 = orig_y1 - l_o

            if y1 < text_height + text_offset_original + 2 * text_spaces:
                y1 = (
                    orig_y1
                    + r_o
                    + text_height
                    + text_offset_original
                    + 2 * text_spaces
                )
                x1 = orig_x1 + r_o

            (text_width, text_height), _ = cv2.getTextSize(
                f" {entity_name}", cv2.FONT_HERSHEY_COMPLEX, text_size, text_line
            )
            text_bg_x1 = x1
            text_bg_y1 = y1 - (
                text_height + text_offset_original + 2 * text_spaces
            )
            text_bg_x2 = x1 + text_width
            text_bg_y2 = y1

            for prev_bbox in previous_bboxes:
                while is_overlapping(
                    (text_bg_x1, text_bg_y1, text_bg_x2, text_bg_y2), prev_bbox
                ):
                    text_bg_y1 += text_height + text_offset_original + 2 * text_spaces
                    text_bg_y2 += text_height + text_offset_original + 2 * text_spaces
                    y1 += text_height + text_offset_original + 2 * text_spaces

                    if text_bg_y2 >= image_h:
                        text_bg_y1 = max(
                            0,
                            image_h
                            - (text_height + text_offset_original + 2 * text_spaces),
                        )
                        text_bg_y2 = image_h
                        y1 = image_h
                        break

            alpha = 0.5
            for i in range(text_bg_y1, text_bg_y2):
                for j in range(text_bg_x1, text_bg_x2):
                    if i < image_h and j < image_w:
                        if j < text_bg_x1 + 1.35 * c_width:
                            bg_color = color
                        else:
                            bg_color = [255, 255, 255]
                        new_image[i, j] = (
                            alpha * new_image[i, j]
                            + (1 - alpha) * np.array(bg_color)
                        ).astype(np.uint8)

            cv2.putText(
                new_image,
                f" {entity_name}",
                (x1, y1 - text_offset_original - 1 * text_spaces),
                cv2.FONT_HERSHEY_COMPLEX,
                text_size,
                (0, 0, 0),
                text_line,
                cv2.LINE_AA,
            )

            previous_bboxes.append(
                (text_bg_x1, text_bg_y1, text_bg_x2, text_bg_y2)
            )

    pil_image = Image.fromarray(new_image[:, :, [2, 1, 0]])

    if save_path:
        pil_image.save(save_path)

    if show:
        pil_image.show()

    return pil_image


ckpt = "microsoft/kosmos-2-patch14-224"
model = AutoModelForVision2Seq.from_pretrained(ckpt).to("cuda")
processor = AutoProcessor.from_pretrained(ckpt)


@spaces.GPU
def generate_predictions(image_input, text_input):
    """
    Generate a grounded image description and annotated entity boxes with Kosmos-2.

    Use this tool when you need to describe an image and identify grounded visual entities.

    Args:
        image_input (PIL.Image.Image): Input image to describe and ground.
        text_input (str): Description mode, either "Brief" or "Detailed".

    Returns:
        tuple: Annotated image, highlighted generated description, and serialized entity data.
    """
    # Save the image and load it again to match the original Kosmos-2 demo.
    # (https://github.com/microsoft/unilm/blob/f4695ed0244a275201fff00bee495f76670fbe70/kosmos-2/demo/gradio_app.py#L345-L346)
    user_image_path = "/tmp/user_input_test_image.jpg"
    image_input.save(user_image_path)

    # This might give different results from the original argument `image_input`
    image_input = Image.open(user_image_path)

    if text_input == "Brief":
        text_input = "An image of"
    elif text_input == "Detailed":
        text_input = "Describe this image in detail:"
    else:
        text_input = f"{text_input}"

    inputs = processor(text=text_input, images=image_input, return_tensors="pt").to(
        "cuda"
    )

    generated_ids = model.generate(
        pixel_values=inputs["pixel_values"],
        input_ids=inputs["input_ids"],
        attention_mask=inputs["attention_mask"],
        image_embeds=None,
        image_embeds_position_mask=inputs["image_embeds_position_mask"],
        use_cache=True,
        max_new_tokens=128,
    )

    generated_text = processor.batch_decode(
        generated_ids, skip_special_tokens=True
    )[0]

    # By default, the generated text is cleanup and the entities are extracted.
    processed_text, entities = processor.post_process_generation(generated_text)

    annotated_image = draw_entity_boxes_on_image(image_input, entities, show=False)

    color_id = -1
    entity_info = []
    filtered_entities = []

    for entity in entities:
        entity_name, (start, end), bboxes = entity

        if start == end:
            # skip bounding bbox without a `phrase` associated
            continue

        color_id += 1
        entity_info.append(((start, end), color_id))
        filtered_entities.append(entity)

    colored_text = []
    prev_start = 0
    end = 0

    for idx, ((start, end), color_id) in enumerate(entity_info):
        if start > prev_start:
            colored_text.append((processed_text[prev_start:start], None))
        colored_text.append((processed_text[start:end], f"{color_id}"))
        prev_start = end

    if end < len(processed_text):
        colored_text.append((processed_text[end : len(processed_text)], None))

    return annotated_image, colored_text, str(filtered_entities)


def main():
    term_of_use = """
### Terms of use

By using this model, users are required to agree to the following terms:

The model is intended for academic and research purposes.
The utilization of the model to create unsuitable material is strictly forbidden and not endorsed by this work.
The accountability for any improper or unacceptable application of the model rests exclusively with the individuals who generated such content.

### License

This project is licensed under the [Microsoft Open Source Code of Conduct](https://opensource.microsoft.com/codeofconduct).
"""

    with gr.Blocks(title="Kosmos-2", theme=gr.themes.Base()).queue() as demo:
        gr.Markdown(
            (
                """
# Kosmos-2: Grounding Multimodal Large Language Models to the World

[[Paper]](https://arxiv.org/abs/2306.14824) [[Code]](https://github.com/microsoft/unilm/blob/master/kosmos-2)
"""
            )
        )

        with gr.Row():
            with gr.Column():
                image_input = gr.Image(type="pil", label="Test Image")
                text_input = gr.Radio(
                    ["Brief", "Detailed"], label="Description Type", value="Brief"
                )
                run_button = gr.Button("Run", visible=True)

            with gr.Column():
                image_output = gr.Image(type="pil")
                text_output1 = gr.HighlightedText(
                    label="Generated Description",
                    combine_adjacent=False,
                    show_legend=True,
                    color_map=color_map,
                )

        with gr.Row():
            with gr.Column():
                gr.Examples(
                    examples=[
                        ["images/two_dogs.jpg", "Detailed"],
                        ["images/snowman.png", "Brief"],
                        ["images/man_ball.png", "Detailed"],
                    ],
                    inputs=[image_input, text_input],
                )

            with gr.Column():
                gr.Examples(
                    examples=[
                        ["images/six_planes.png", "Brief"],
                        ["images/quadrocopter.jpg", "Brief"],
                        ["images/carnaby_street.jpg", "Brief"],
                    ],
                    inputs=[image_input, text_input],
                )

        gr.Markdown(term_of_use)

        # record which text span (label) is selected
        selected = gr.Number(-1, show_label=False, visible=False)

        # record the current `entities`
        entity_output = gr.Textbox(visible=False)

        # get the current selected span label
        def get_text_span_label(evt: gr.SelectData):
            if evt.value[-1] is None:
                return -1

            return int(evt.value[-1])

        # and set this information to `selected`
        text_output1.select(
            get_text_span_label,
            None,
            selected,
            api_visibility="private",
        )

        # update output image when we change the span (enity) selection
        def update_output_image(img_input, image_output, entities, idx):
            entities = ast.literal_eval(entities)
            updated_image = draw_entity_boxes_on_image(
                img_input, entities, entity_index=idx
            )
            return updated_image

        selected.change(
            update_output_image,
            [image_input, image_output, entity_output, selected],
            [image_output],
            api_visibility="private",
        )

        run_button.click(
            fn=generate_predictions,
            inputs=[image_input, text_input],
            outputs=[image_output, text_output1, entity_output],
            show_progress=True,
            queue=True,
        )

    demo.launch(share=False, ssr_mode=False, mcp_server=True)


if __name__ == "__main__":
    main()
# trigger