File size: 12,991 Bytes
ed8c5ad
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
106ce32
 
ed8c5ad
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
106ce32
 
 
ed8c5ad
6d7c8da
106ce32
ed8c5ad
 
 
 
 
 
 
 
 
 
 
 
 
 
 
106ce32
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
ed8c5ad
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
6d7c8da
ed8c5ad
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
262ca90
ed8c5ad
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
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
416
417
418
419
420
421
422
423
424
425
426
427
428
429
430
431
432
433
434
435
436
437
438
439
440
441
442
443
444
445
446
447
448
449
450
451
452
453
454
455
456
457
458
459
460
461
462
463
464
465
466
467
468
469
470
471
472
473
import csv
import json
import os
import pickle
import random
import string
import sys
import time
from glob import glob

import datasets
import gdown
import gradio as gr
import matplotlib.pyplot as plt
import numpy as np
import pandas as pd
import torch
import torchvision
from huggingface_hub import HfApi, login, snapshot_download
from PIL import Image
import re
from fnmatch import translate

session_token = os.environ.get("SessionToken")
login(token=session_token)

csv.field_size_limit(sys.maxsize)

np.random.seed(int(time.time()))

with open("./imagenet_hard_nearest_indices.pkl", "rb") as f:
    knn_results = pickle.load(f)

with open("imagenet-labels.json") as f:
    wnid_to_label = json.load(f)

with open("id_to_label.json", "r") as f:
    id_to_labels = json.load(f)

imagenet_training_samples_path = "imagenet_samples"

bad_items = open("./ex2.txt", "r").read().split("\n")
bad_items = [x.split(".")[0] for x in bad_items]
bad_items = [int(x) for x in bad_items if x != ""]

NUMBER_OF_IMAGES = len(bad_items)


gdown.cached_download(
    url="https://huggingface.co/datasets/taesiri/imagenet_hard_review_samples/resolve/main/data.zip",
    path="./data.zip",
    quiet=False,
    md5="ece2720fed664e71799f316a881d4324",
)

# EXTRACT if needed

if not os.path.exists("./imagenet_samples") or not os.path.exists(
    "./knn_cache_for_imagenet_hard"
):
    torchvision.datasets.utils.extract_archive(
        from_path="data.zip",
        to_path="./",
        remove_finished=False,
    )

imagenet_hard = datasets.load_dataset("taesiri/imagenet-hard", split="validation")

def update_snapshot(username):
    escaped_username = re.escape(username)
    pattern = f"*{escaped_username}*.json"

    output_dir = snapshot_download(
        repo_id="taesiri/imagenet_hard_review_data_r2",
        allow_patterns=translate(pattern),
        repo_type="dataset",
    )
    files = glob(f"{output_dir}/*.json")

    df = pd.DataFrame()
    columns = ["id", "user_id", "time", "decision"]
    rows = []
    for file in files:
        with open(file) as f:
            data = json.load(f)
            tdf = [data[x] for x in columns]
            rows.append(tdf)

    df = pd.DataFrame(rows, columns=columns)

    # download and append all CSV files
    output_dir = snapshot_download(
        repo_id="taesiri/imagenet_hard_review_data_r3",
        allow_patterns="*.csv",
        repo_type="dataset",
    )
    files = glob(f"{output_dir}/*.csv")

    if len(files) > 0:
        csv_dataframes = [pd.read_csv(file) for file in files]
        csv_dataframes = pd.concat(csv_dataframes, ignore_index=True)
        df = pd.concat([df, csv_dataframes], ignore_index=True)

    # remove duplicate rows
    df = df.drop_duplicates(subset=["id", "user_id"], keep="last")
    df = df[df["user_id"] == username]
    return df


def generate_dataset(username):
    global NUMBER_OF_IMAGES
    df = update_snapshot(username)

    all_images = set(bad_items)
    answered = set(df.id)
    remaining = list(all_images - answered)
    # shuffle remaining
    random.shuffle(remaining)

    NUMBER_OF_IMAGES = len(bad_items)

    print(f"NUMBER_OF_IMAGES: {NUMBER_OF_IMAGES}")
    print(f"Remaining: {len(remaining)}")

    if NUMBER_OF_IMAGES == 0:
        return []

    data = []
    for i, image in enumerate(remaining):
        data.append(
            {
                "id": remaining[i],
            }
        )
    return data


def string_to_image(text):
    text = text.replace("_", " ").lower().replace(", ", "\n")
    # Create a blank white square image
    img = np.ones((220, 75, 3))

    fig, ax = plt.subplots(figsize=(6, 2.25))
    ax.imshow(img, extent=[0, 1, 0, 1])
    ax.text(0.5, 0.75, text, fontsize=18, ha="center", va="center")
    ax.set_xticks([])
    ax.set_yticks([])
    ax.set_xticklabels([])
    ax.set_yticklabels([])
    for spine in ax.spines.values():
        spine.set_visible(False)

    return fig


all_samples = glob("./imagenet_samples/*.JPEG")
qid_to_sample = {
    int(x.split("/")[-1].split(".")[0].split("_")[0]): x for x in all_samples
}


def get_training_samples(qid):
    labels_id = imagenet_hard[int(qid)]["label"]
    samples = [qid_to_sample[x] for x in labels_id]
    return samples


def load_sample(data, current_index):
    image_id = data[current_index]["id"]
    qimage = imagenet_hard[int(image_id)]["image"]
    # labels = data[current_index]["correct_label"]
    labels = imagenet_hard[int(image_id)]["english_label"]
    # print(f"Image ID: {image_id}")
    # print(f"Labels: {labels}")

    return qimage, labels


def preprocessing(data, current_index, history, username):
    data = generate_dataset(username)

    remaining_images = len(data)
    labeled_images = len(bad_items) - remaining_images

    if remaining_images == 0:
        fake_plot = string_to_image("No more images to review")
        empty_image = Image.new("RGB", (224, 224))
        return (
            empty_image,
            fake_plot,
            current_index,
            history,
            data,
            None,
            labeled_images,
        )

    current_index = 0
    qimage, labels = load_sample(data, current_index)
    image_id = data[current_index]["id"]
    training_samples_image = get_training_samples(image_id)
    training_samples_image = [
        Image.open(x).convert("RGB") for x in training_samples_image
    ]

    # labels is a list of labels, conver it to a string
    labels = ", ".join(labels)
    label_plot = string_to_image(labels)

    return (
        qimage,
        label_plot,
        current_index,
        history,
        data,
        training_samples_image,
        labeled_images,
    )


def update_app(decision, data, current_index, history, username):
    global NUMBER_OF_IMAGES
    if current_index == -1:
        fake_plot = string_to_image("Please Enter your username and load samples")
        empty_image = Image.new("RGB", (224, 224))
        return empty_image, fake_plot, current_index, history, data, None, 0

    if current_index == NUMBER_OF_IMAGES - 1:
        time_stamp = int(time.time())

        image_id = data[current_index]["id"]
        # convert to percentage
        dicision_dict = {
            "id": int(image_id),
            "user_id": username,
            "time": time_stamp,
            "decision": decision,
        }

        # upload the decision to the server
        temp_filename = f"results_{username}_{time_stamp}.json"
        # convert decision_dict to json and save it on the disk
        with open(temp_filename, "w") as f:
            json.dump(dicision_dict, f)

        api = HfApi()
        api.upload_file(
            path_or_fileobj=temp_filename,
            path_in_repo=temp_filename,
            repo_id="taesiri/imagenet_hard_review_data_r2",
            repo_type="dataset",
        )

        os.remove(temp_filename)

        fake_plot = string_to_image("Thank you for your time!")
        empty_image = Image.new("RGB", (224, 224))

        remaining_images = len(data)
        labeled_images = (len(bad_items) - remaining_images) + current_index

        return (
            empty_image,
            fake_plot,
            current_index,
            history,
            data,
            None,
            labeled_images + 1,
        )

    if current_index >= 0 and current_index < NUMBER_OF_IMAGES - 1:
        time_stamp = int(time.time())

        image_id = data[current_index]["id"]
        # convert to percentage
        dicision_dict = {
            "id": int(image_id),
            "user_id": username,
            "time": time_stamp,
            "decision": decision,
        }

        # upload the decision to the server
        temp_filename = f"results_{username}_{time_stamp}.json"
        # convert decision_dict to json and save it on the disk
        with open(temp_filename, "w") as f:
            json.dump(dicision_dict, f)

        api = HfApi()
        api.upload_file(
            path_or_fileobj=temp_filename,
            path_in_repo=temp_filename,
            repo_id="taesiri/imagenet_hard_review_data_r2",
            repo_type="dataset",
        )

        os.remove(temp_filename)

        # Load the Next Image

        current_index += 1
        qimage, labels = load_sample(data, current_index)
        image_id = data[current_index]["id"]
        training_samples_image = get_training_samples(image_id)
        training_samples_image = [
            Image.open(x).convert("RGB") for x in training_samples_image
        ]

        # labels is a list of labels, conver it to a string
        labels = ", ".join(labels)
        label_plot = string_to_image(labels)

        remaining_images = len(data)
        labeled_images = (len(bad_items) - remaining_images) + current_index

        return (
            qimage,
            label_plot,
            current_index,
            history,
            data,
            training_samples_image,
            labeled_images,
        )


newcss = """
#query_image{
}

#nn_gallery {
  height: auto !important;
}

#sample_gallery {
    height: auto !important;
}


/* Set display to flex for the parent element */
.svelte-parentrowclass {
  display: flex;
}

/* Set the flex-grow property for the children elements */
.svelte-parentrowclass > #query_image {
  min-width: min(400px, 40%);
  flex : 1;
  flex-grow: 0; !important;
  border-style: solid;
  height: auto !important;
}

.svelte-parentrowclass > .svelte-rightcolumn {
  flex: 2;
  flex-grow: 0; !important;
  min-width: min(600px, 60%);
}



"""

with gr.Blocks(css=newcss, theme=gr.themes.Soft()) as demo:
    data_gr = gr.State({})
    current_index = gr.State(-1)
    history = gr.State({})

    gr.Markdown("# Help Us to Clean `ImageNet-Hard`!")

    gr.Markdown("## Instructions")
    gr.Markdown(
        "Please enter your username and press `Load Samples`. The loading process might take up to a minute. Once the loading is done, you can start reviewing the samples."
    )
    gr.Markdown(
        """For each image, please select one of the following options: `Accept`, `Not Sure!`, `Reject`.
        - If you think any of the labels are correct, please select `Accept`.
        - If you think none of the labels matching the image, please select `Reject`. 
        - If you are not sure about the label, please select `Not Sure!`. 

        You can refer to `Training samples` if you are not sure about the target label.
        """
    )

    random_str = "".join(
        random.choice(string.ascii_lowercase + string.digits) for _ in range(5)
    )

    with gr.Column():
        with gr.Row():
            username = gr.Textbox(label="Username", value=f"user-{random_str}")
            labeled_images = gr.Textbox(label="Labeled Images", value="0")
            total_images = gr.Textbox(label="Total Images", value=len(bad_items))

        prepare_btn = gr.Button(value="Load Samples")

    with gr.Column():
        with gr.Row():
            accept_btn = gr.Button(value="Accept")
            myabe_btn = gr.Button(value="Not Sure!")
            reject_btn = gr.Button(value="Reject")
        with gr.Row(elem_id="parent_row", elem_classes="svelte-parentrowclass"):
            query_image = gr.Image(type="pil", label="Query", elem_id="query_image")
            with gr.Column(
                elem_id="samples_col",
                elem_classes="svelte-rightcolumn",
            ):
                label_plot = gr.Plot(
                    label="Is this a correct label for this image?", type="fig"
                )
                training_samples = gr.Gallery(
                    type="pil", label="Training samples", elem_id="sample_gallery"
                )

    accept_btn.click(
        update_app,
        inputs=[accept_btn, data_gr, current_index, history, username],
        outputs=[
            query_image,
            label_plot,
            current_index,
            history,
            data_gr,
            training_samples,
            labeled_images,
        ],
    )
    myabe_btn.click(
        update_app,
        inputs=[myabe_btn, data_gr, current_index, history, username],
        outputs=[
            query_image,
            label_plot,
            current_index,
            history,
            data_gr,
            training_samples,
            labeled_images,
        ],
    )

    reject_btn.click(
        update_app,
        inputs=[reject_btn, data_gr, current_index, history, username],
        outputs=[
            query_image,
            label_plot,
            current_index,
            history,
            data_gr,
            training_samples,
            labeled_images,
        ],
    )

    prepare_btn.click(
        preprocessing,
        inputs=[data_gr, current_index, history, username],
        outputs=[
            query_image,
            label_plot,
            current_index,
            history,
            data_gr,
            training_samples,
            labeled_images,
        ],
    )


demo.launch(debug=False)