File size: 12,878 Bytes
b793f0c
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
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
import argparse
import os
import copy

import numpy as np
import json
import torch
import torchvision
from PIL import Image, ImageDraw, ImageFont
import litellm

# Grounding DINO
import GroundingDINO.groundingdino.datasets.transforms as T
from GroundingDINO.groundingdino.models import build_model
from GroundingDINO.groundingdino.util import box_ops
from GroundingDINO.groundingdino.util.slconfig import SLConfig
from GroundingDINO.groundingdino.util.utils import clean_state_dict, get_phrases_from_posmap

# segment anything
from segment_anything import build_sam, SamPredictor 
import cv2
import numpy as np
import matplotlib.pyplot as plt

# Tag2Text
from ram.models import tag2text_caption
from ram import inference_tag2text
import torchvision.transforms as TS

# ChatGPT or nltk is required when using captions
# import openai
# import nltk

def load_image(image_path):
    # load image
    image_pil = Image.open(image_path).convert("RGB")  # load image

    transform = T.Compose(
        [
            T.RandomResize([800], max_size=1333),
            T.ToTensor(),
            T.Normalize([0.485, 0.456, 0.406], [0.229, 0.224, 0.225]),
        ]
    )
    image, _ = transform(image_pil, None)  # 3, h, w
    return image_pil, image


def generate_caption(raw_image, device):
    # unconditional image captioning
    if device == "cuda":
        inputs = processor(raw_image, return_tensors="pt").to("cuda", torch.float16)
    else:
        inputs = processor(raw_image, return_tensors="pt")
    out = blip_model.generate(**inputs)
    caption = processor.decode(out[0], skip_special_tokens=True)
    return caption


def generate_tags(caption, split=',', max_tokens=100, model="gpt-3.5-turbo"):
    lemma = nltk.wordnet.WordNetLemmatizer()
    if openai_key:
        prompt = [
            {
                'role': 'system',
                'content': 'Extract the unique nouns in the caption. Remove all the adjectives. ' + \
                           f'List the nouns in singular form. Split them by "{split} ". ' + \
                           f'Caption: {caption}.'
            }
        ]
        response = litellm.completion(model=model, messages=prompt, temperature=0.6, max_tokens=max_tokens)
        reply = response['choices'][0]['message']['content']
        # sometimes return with "noun: xxx, xxx, xxx"
        tags = reply.split(':')[-1].strip()
    else:
        nltk.download(['punkt', 'averaged_perceptron_tagger', 'wordnet'])
        tags_list = [word for (word, pos) in nltk.pos_tag(nltk.word_tokenize(caption)) if pos[0] == 'N']
        tags_lemma = [lemma.lemmatize(w) for w in tags_list]
        tags = ', '.join(map(str, tags_lemma))
    return tags


def check_caption(caption, pred_phrases, max_tokens=100, model="gpt-3.5-turbo"):
    object_list = [obj.split('(')[0] for obj in pred_phrases]
    object_num = []
    for obj in set(object_list):
        object_num.append(f'{object_list.count(obj)} {obj}')
    object_num = ', '.join(object_num)
    print(f"Correct object number: {object_num}")

    if openai_key:
        prompt = [
            {
                'role': 'system',
                'content': 'Revise the number in the caption if it is wrong. ' + \
                           f'Caption: {caption}. ' + \
                           f'True object number: {object_num}. ' + \
                           'Only give the revised caption: '
            }
        ]
        response = litellm.completion(model=model, messages=prompt, temperature=0.6, max_tokens=max_tokens)
        reply = response['choices'][0]['message']['content']
        # sometimes return with "Caption: xxx, xxx, xxx"
        caption = reply.split(':')[-1].strip()
    return caption


def load_model(model_config_path, model_checkpoint_path, device):
    args = SLConfig.fromfile(model_config_path)
    args.device = device
    model = build_model(args)
    checkpoint = torch.load(model_checkpoint_path, map_location="cpu")
    load_res = model.load_state_dict(clean_state_dict(checkpoint["model"]), strict=False)
    print(load_res)
    _ = model.eval()
    return model


def get_grounding_output(model, image, caption, box_threshold, text_threshold,device="cpu"):
    caption = caption.lower()
    caption = caption.strip()
    if not caption.endswith("."):
        caption = caption + "."
    model = model.to(device)
    image = image.to(device)
    with torch.no_grad():
        outputs = model(image[None], captions=[caption])
    logits = outputs["pred_logits"].cpu().sigmoid()[0]  # (nq, 256)
    boxes = outputs["pred_boxes"].cpu()[0]  # (nq, 4)
    logits.shape[0]

    # filter output
    logits_filt = logits.clone()
    boxes_filt = boxes.clone()
    filt_mask = logits_filt.max(dim=1)[0] > box_threshold
    logits_filt = logits_filt[filt_mask]  # num_filt, 256
    boxes_filt = boxes_filt[filt_mask]  # num_filt, 4
    logits_filt.shape[0]

    # get phrase
    tokenlizer = model.tokenizer
    tokenized = tokenlizer(caption)
    # build pred
    pred_phrases = []
    scores = []
    for logit, box in zip(logits_filt, boxes_filt):
        pred_phrase = get_phrases_from_posmap(logit > text_threshold, tokenized, tokenlizer)
        pred_phrases.append(pred_phrase + f"({str(logit.max().item())[:4]})")
        scores.append(logit.max().item())

    return boxes_filt, torch.Tensor(scores), pred_phrases


def show_mask(mask, ax, random_color=False):
    if random_color:
        color = np.concatenate([np.random.random(3), np.array([0.6])], axis=0)
    else:
        color = np.array([30/255, 144/255, 255/255, 0.6])
    h, w = mask.shape[-2:]
    mask_image = mask.reshape(h, w, 1) * color.reshape(1, 1, -1)
    ax.imshow(mask_image)


def show_box(box, ax, label):
    x0, y0 = box[0], box[1]
    w, h = box[2] - box[0], box[3] - box[1]
    ax.add_patch(plt.Rectangle((x0, y0), w, h, edgecolor='green', facecolor=(0,0,0,0), lw=2)) 
    ax.text(x0, y0, label)


def save_mask_data(output_dir, caption, mask_list, box_list, label_list):
    value = 0  # 0 for background

    mask_img = torch.zeros(mask_list.shape[-2:])
    for idx, mask in enumerate(mask_list):
        mask_img[mask.cpu().numpy()[0] == True] = value + idx + 1
    plt.figure(figsize=(10, 10))
    plt.imshow(mask_img.numpy())
    plt.axis('off')
    plt.savefig(os.path.join(output_dir, 'mask.jpg'), bbox_inches="tight", dpi=300, pad_inches=0.0)

    json_data = {
        'caption': caption,
        'mask':[{
            'value': value,
            'label': 'background'
        }]
    }
    for label, box in zip(label_list, box_list):
        value += 1
        name, logit = label.split('(')
        logit = logit[:-1] # the last is ')'
        json_data['mask'].append({
            'value': value,
            'label': name,
            'logit': float(logit),
            'box': box.numpy().tolist(),
        })
    with open(os.path.join(output_dir, 'label.json'), 'w') as f:
        json.dump(json_data, f)
    

if __name__ == "__main__":

    parser = argparse.ArgumentParser("Grounded-Segment-Anything Demo", add_help=True)
    parser.add_argument("--config", type=str, required=True, help="path to config file")
    parser.add_argument(
        "--tag2text_checkpoint", type=str, required=True, help="path to checkpoint file"
    )
    parser.add_argument(
        "--grounded_checkpoint", type=str, required=True, help="path to checkpoint file"
    )
    parser.add_argument(
        "--sam_checkpoint", type=str, required=True, help="path to checkpoint file"
    )
    parser.add_argument("--input_image", type=str, required=True, help="path to image file")
    parser.add_argument("--split", default=",", type=str, help="split for text prompt")
    parser.add_argument("--openai_key", type=str, help="key for chatgpt")
    parser.add_argument("--openai_proxy", default=None, type=str, help="proxy for chatgpt")
    parser.add_argument(
        "--output_dir", "-o", type=str, default="outputs", required=True, help="output directory"
    )

    parser.add_argument("--box_threshold", type=float, default=0.25, help="box threshold")
    parser.add_argument("--text_threshold", type=float, default=0.2, help="text threshold")
    parser.add_argument("--iou_threshold", type=float, default=0.5, help="iou threshold")

    parser.add_argument("--device", type=str, default="cpu", help="running on cpu only!, default=False")
    args = parser.parse_args()

    # cfg
    config_file = args.config  # change the path of the model config file
    tag2text_checkpoint = args.tag2text_checkpoint  # change the path of the model
    grounded_checkpoint = args.grounded_checkpoint  # change the path of the model
    sam_checkpoint = args.sam_checkpoint
    image_path = args.input_image
    split = args.split
    openai_key = args.openai_key
    openai_proxy = args.openai_proxy
    output_dir = args.output_dir
    box_threshold = args.box_threshold
    text_threshold = args.text_threshold
    iou_threshold = args.iou_threshold
    device = args.device
    
    # ChatGPT or nltk is required when using captions
    # openai.api_key = openai_key
    # if openai_proxy:
        # openai.proxy = {"http": openai_proxy, "https": openai_proxy}

    # make dir
    os.makedirs(output_dir, exist_ok=True)
    # load image
    image_pil, image = load_image(image_path)
    # load model
    model = load_model(config_file, grounded_checkpoint, device=device)

    # visualize raw image
    image_pil.save(os.path.join(output_dir, "raw_image.jpg"))

    # initialize Tag2Text
    normalize = TS.Normalize(mean=[0.485, 0.456, 0.406],
                                     std=[0.229, 0.224, 0.225])
    transform = TS.Compose([
                    TS.Resize((384, 384)),
                    TS.ToTensor(), normalize
                ])
    
    # filter out attributes and action categories which are difficult to grounding
    delete_tag_index = []
    for i in range(3012, 3429):
        delete_tag_index.append(i)

    specified_tags='None'
    # load model
    tag2text_model = tag2text_caption(pretrained=tag2text_checkpoint,
                                        image_size=384,
                                        vit='swin_b',
                                        delete_tag_index=delete_tag_index)
    # threshold for tagging
    # we reduce the threshold to obtain more tags
    tag2text_model.threshold = 0.64 
    tag2text_model.eval()

    tag2text_model = tag2text_model.to(device)
    raw_image = image_pil.resize(
                    (384, 384))
    raw_image  = transform(raw_image).unsqueeze(0).to(device)

    res = inference_tag2text(raw_image , tag2text_model, specified_tags)

    # Currently ", " is better for detecting single tags
    # while ". " is a little worse in some case
    text_prompt=res[0].replace(' |', ',')
    caption=res[2]

    print(f"Caption: {caption}")
    print(f"Tags: {text_prompt}")

    # run grounding dino model
    boxes_filt, scores, pred_phrases = get_grounding_output(
        model, image, text_prompt, box_threshold, text_threshold, device=device
    )

    # initialize SAM
    predictor = SamPredictor(build_sam(checkpoint=sam_checkpoint).to(device))
    image = cv2.imread(image_path)
    image = cv2.cvtColor(image, cv2.COLOR_BGR2RGB)
    predictor.set_image(image)

    size = image_pil.size
    H, W = size[1], size[0]
    for i in range(boxes_filt.size(0)):
        boxes_filt[i] = boxes_filt[i] * torch.Tensor([W, H, W, H])
        boxes_filt[i][:2] -= boxes_filt[i][2:] / 2
        boxes_filt[i][2:] += boxes_filt[i][:2]

    boxes_filt = boxes_filt.cpu()
    # use NMS to handle overlapped boxes
    print(f"Before NMS: {boxes_filt.shape[0]} boxes")
    nms_idx = torchvision.ops.nms(boxes_filt, scores, iou_threshold).numpy().tolist()
    boxes_filt = boxes_filt[nms_idx]
    pred_phrases = [pred_phrases[idx] for idx in nms_idx]
    print(f"After NMS: {boxes_filt.shape[0]} boxes")
    caption = check_caption(caption, pred_phrases)
    print(f"Revise caption with number: {caption}")

    transformed_boxes = predictor.transform.apply_boxes_torch(boxes_filt, image.shape[:2]).to(device)

    masks, _, _ = predictor.predict_torch(
        point_coords = None,
        point_labels = None,
        boxes = transformed_boxes.to(device),
        multimask_output = False,
    )
    
    # draw output image
    plt.figure(figsize=(10, 10))
    plt.imshow(image)
    for mask in masks:
        show_mask(mask.cpu().numpy(), plt.gca(), random_color=True)
    for box, label in zip(boxes_filt, pred_phrases):
        show_box(box.numpy(), plt.gca(), label)

    plt.title('Tag2Text-Captioning: ' + caption + '\n' + 'Tag2Text-Tagging' + text_prompt + '\n')
    plt.axis('off')
    plt.savefig(
        os.path.join(output_dir, "automatic_label_output.jpg"), 
        bbox_inches="tight", dpi=300, pad_inches=0.0
    )

    save_mask_data(output_dir, caption, masks, boxes_filt, pred_phrases)