Analyzing Object Detection Dataset¶

In this tutorial, we will analyze an object detection dataset with bounding boxes and identify potential issues. By the end of the notebook, you'll discover how to load COCO-format bounding box annotations into fastdup and inspect the dataset for issues at the bounding box level.

Installation¶

In [ ]:

!pip install -Uq fastdup plotly gdown

Now, test the installation by printing out the version. If there's no error message, we are ready to go!

In [1]:

import fastdup
fastdup.__version__

/usr/bin/dpkg

Out[1]:

'1.33'

Download Dataset¶

We will be using the COCO minitrain dataset for this tutorial. COCO minitrain is a curated mini training set with about 25,000 images, or 20% of the original COCO dataset.

Let's download the dataset into our local drive.

In [ ]:

# Download coco minitrain dataset.
!gdown --fuzzy https://drive.google.com/file/d/1iSXVTlkV1_DhdYpVDqsjlT4NJFQ7OkyK/view
!unzip -qq coco_minitrain_25k.zip

# Download csv annotations
!cd coco_minitrain_25k/annotations && gdown --fuzzy https://drive.google.com/file/d/1i12p23cXlqp1QrXjAD_vu467r4q67Mq9/view

Load Annotations¶

fastdup requires the annotations to be in a specific format.

We will use a simple converter to convert the COCO format JSON annotation file into the fastdup annotation dataframe. This converter is applicable to any dataset that uses the COCO format.

In [2]:

import pandas as pd

In [3]:

coco_csv = 'coco_minitrain_25k/annotations/coco_minitrain2017.csv'
coco_annotations = pd.read_csv(coco_csv, 
                               header=None, 
                               names=['filename', 'col_x', 'row_y',
                                      'width', 'height', 'label', 'ext'])

coco_annotations['split'] = 'train'  # Only train files were loaded
coco_annotations['filename'] = coco_annotations['filename'].apply(lambda x: 'images/train2017/'+x)
coco_annotations = coco_annotations.drop_duplicates()

In [4]:

coco_annotations.head(3)

Out[4]:

	filename	col_x	row_y	width	height	label	split
0	images/train2017/000000131075.jpg	20.23	55.98	313.49	326.50	tv	train
1	images/train2017/000000131075.jpg	176.90	381.12	286.20	136.63	laptop	train
2	images/train2017/000000131075.jpg	369.96	361.35	72.76	73.91	laptop	train

Run fastdup¶

Run fastdup with annotations on the dataset. If you're running on a free Google Colab instance, you may want to specify num_images to limit the run to fewer images.

In [ ]:

fd = fastdup.create(input_dir='./coco_minitrain_25k')
fd.run(annotations=coco_annotations, ccthreshold=0.9)

Class Distribution¶

The dataset contains 25k images and 183k objects, for an average of 7.3 objects per image.

Interestingly, we see a highly unbalanced class distribution. All 80 COCO classes are present here, but the distribution of classes is strongly skewed towards the person class, which accounts for over 56k instances (30.6%). Car and Chair classes also contain over 8k instances each, while the toaster and hair drier classes contain as few as 40 instances.

Using Plotly we get a useful interactive histogram.

In [6]:

import plotly.express as px
fig = px.histogram(coco_annotations, x="label")
fig.show()

Outliers¶

Using fastdup's gallery feature, we can visualize outliers from the run.

In [8]:

fd.vis.outliers_gallery()

100%|███████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████| 20/20 [00:00<00:00, 23484.34it/s]

Stored outliers visual view in  work_dir/galleries/outliers.html
########################################################################################
Would you like to see awesome visualizations for some of the most popular academic datasets?
Click here to see and learn more: https://app.visual-layer.com/vl-datasets?utm_source=fastdup
########################################################################################

Outliers Report

Showing image outliers, one per row

Info
Distance	0.900001
Path	/crops/imagestrain2017000000171809.jpg_79_96_368_213.jpg
label	cat

Info
Distance	0.900001
Path	/crops/imagestrain2017000000171809.jpg_0_17_508_440.jpg
label	bench

Info
Distance	0.900002
Path	/crops/imagestrain2017000000579267.jpg_126_123_30_83.jpg
label	person

Info
Distance	0.900002
Path	/crops/imagestrain2017000000197155.jpg_598_257_14_16.jpg
label	dining table

Info
Distance	0.900002
Path	/crops/imagestrain2017000000080140.jpg_46_156_25_72.jpg
label	person

Info
Distance	0.900003
Path	/crops/imagestrain2017000000538879.jpg_59_332_15_30.jpg
label	traffic light

Info
Distance	0.900003
Path	/crops/imagestrain2017000000189943.jpg_254_52_18_31.jpg
label	snowboard

Info
Distance	0.900004
Path	/crops/imagestrain2017000000511977.jpg_1_258_170_24.jpg
label	bench

Info
Distance	0.900004
Path	/crops/imagestrain2017000000033687.jpg_104_295_23_32.jpg
label	fire hydrant

Info
Distance	0.900004
Path	/crops/imagestrain2017000000511977.jpg_1_304_205_28.jpg
label	bench

Info
Distance	0.900005
Path	/crops/imagestrain2017000000388281.jpg_25_153_36_23.jpg
label	tv

Info
Distance	0.900005
Path	/crops/imagestrain2017000000169947.jpg_143_272_353_166.jpg
label	elephant

Info
Distance	0.900005
Path	/crops/imagestrain2017000000296901.jpg_337_239_23_18.jpg
label	tv

Info
Distance	0.900008
Path	/crops/imagestrain2017000000123177.jpg_15_233_16_20.jpg
label	kite

Info
Distance	0.900008
Path	/crops/imagestrain2017000000396519.jpg_511_166_15_57.jpg
label	person

Info
Distance	0.900008
Path	/crops/imagestrain2017000000517579.jpg_244_245_103_269.jpg
label	horse

Info
Distance	0.900008
Path	/crops/imagestrain2017000000517579.jpg_236_195_116_177.jpg
label	person

Info
Distance	0.900008
Path	/crops/imagestrain2017000000571768.jpg_259_131_15_69.jpg
label	person

Info
Distance	0.900011
Path	/crops/imagestrain2017000000578652.jpg_0_73_282_270.jpg
label	banana

Info
Distance	0.900011
Path	/crops/imagestrain2017000000392391.jpg_397_175_48_17.jpg
label	car

Out[8]:

Size and Shape Issues¶

Objects come in various shapes and sizes, and sometimes objects might be incorrectly labeled, or too small to be useful. We will now find the smallest, narrowest and widest objects, and assess their usefulness.

In [9]:

fd.annotations()

Out[9]:

	col_x	row_y	width	height	label	ext	split	crop_filename	filename	index	is_valid
0	20.23	55.98	313.49	326.50	tv	0	train	work_dir/crops/coco_minitrain_25kimagestrain2017000000131075.jpg_20_55_313_326.jpg	coco_minitrain_25k/images/train2017/000000131075.jpg	0.0	True
1	176.90	381.12	286.20	136.63	laptop	0	train	work_dir/crops/coco_minitrain_25kimagestrain2017000000131075.jpg_176_381_286_136.jpg	coco_minitrain_25k/images/train2017/000000131075.jpg	1.0	True
2	369.96	361.35	72.76	73.91	laptop	0	train	work_dir/crops/coco_minitrain_25kimagestrain2017000000131075.jpg_369_361_72_73.jpg	coco_minitrain_25k/images/train2017/000000131075.jpg	2.0	True
3	411.68	417.87	66.32	129.44	chair	0	train	work_dir/crops/coco_minitrain_25kimagestrain2017000000131075.jpg_411_417_66_129.jpg	coco_minitrain_25k/images/train2017/000000131075.jpg	3.0	True
4	367.31	363.25	72.27	67.01	tv	0	train	work_dir/crops/coco_minitrain_25kimagestrain2017000000131075.jpg_367_363_72_67.jpg	coco_minitrain_25k/images/train2017/000000131075.jpg	4.0	True
...	...	...	...	...	...	...	...	...	...	...	...
183557	2.45	0.91	94.03	181.51	car	0	train	work_dir/crops/coco_minitrain_25kimagestrain2017000000262103.jpg_2_0_94_181.jpg	coco_minitrain_25k/images/train2017/000000262103.jpg	164947.0	True
183558	6.10	214.53	331.31	262.83	boat	0	train	work_dir/crops/coco_minitrain_25kimagestrain2017000000393195.jpg_6_214_331_262.jpg	coco_minitrain_25k/images/train2017/000000393195.jpg	164948.0	True
183559	46.37	3.34	593.63	478.66	person	0	train	work_dir/crops/coco_minitrain_25kimagestrain2017000000393195.jpg_46_3_593_478.jpg	coco_minitrain_25k/images/train2017/000000393195.jpg	164949.0	True
183560	419.40	0.88	217.84	309.23	person	0	train	work_dir/crops/coco_minitrain_25kimagestrain2017000000393195.jpg_419_0_217_309.jpg	coco_minitrain_25k/images/train2017/000000393195.jpg	164950.0	True
183561	4.21	1.17	628.93	421.75	fire hydrant	0	train	work_dir/crops/coco_minitrain_25kimagestrain2017000000131067.jpg_4_1_628_421.jpg	coco_minitrain_25k/images/train2017/000000131067.jpg	164951.0	True

164970 rows × 11 columns

In [10]:

annot = fd.annotations()
annot['area'] = annot['width'] * annot['height']
annot['aspect'] = annot['width'] / annot['height']

/tmp/ipykernel_997551/2907564842.py:2: SettingWithCopyWarning:


A value is trying to be set on a copy of a slice from a DataFrame.
Try using .loc[row_indexer,col_indexer] = value instead

See the caveats in the documentation: https://pandas.pydata.org/pandas-docs/stable/user_guide/indexing.html#returning-a-view-versus-a-copy

/tmp/ipykernel_997551/2907564842.py:3: SettingWithCopyWarning:


A value is trying to be set on a copy of a slice from a DataFrame.
Try using .loc[row_indexer,col_indexer] = value instead

See the caveats in the documentation: https://pandas.pydata.org/pandas-docs/stable/user_guide/indexing.html#returning-a-view-versus-a-copy

In [11]:

# Smallest 5% of objects:
smallest_objects = annot[annot['area'] < annot['area'].quantile(0.05)].sort_values(by=['area'])

# 5% of extreme aspect ratios
aspect_ratio_objects = annot[(annot['aspect'] < annot['aspect'].quantile(0.05))
                             | (annot['aspect'] > annot['aspect'].quantile(0.95))].sort_values(by=['aspect'])

Let's see the smallest objects labeled in the dataset.

In [12]:

smallest_objects.head(3)

Out[12]:

	col_x	row_y	width	height	label	split	crop_filename	filename	index	is_valid	area	aspect
110332	580.76	232.51	10.00	10.00	person	train	work_dir/crops/coco_minitrain_25kimagestrain2017000000205392.jpg_580_232_10_10.jpg	coco_minitrain_25k/images/train2017/000000205392.jpg	99244.0	True	100.0000	1.0
131669	183.84	300.01	10.02	10.02	apple	train	work_dir/crops/coco_minitrain_25kimagestrain2017000000320957.jpg_183_300_10_10.jpg	coco_minitrain_25k/images/train2017/000000320957.jpg	118354.0	True	100.4004	1.0
21380	301.32	175.41	10.06	10.06	person	train	work_dir/crops/coco_minitrain_25kimagestrain2017000000407413.jpg_301_175_10_10.jpg	coco_minitrain_25k/images/train2017/000000407413.jpg	19248.0	True	101.2036	1.0

Let's visualize what the 3 smallest images look like.

The following image is labeled as a person in the dataset.

In [13]:

from IPython.display import Image
Image(filename="work_dir/crops/coco_minitrain_25kimagestrain2017000000205392.jpg_580_232_10_10.jpg")

Out[13]:

The following image is labeled as an apple in the dataset.

In [14]:

Image(filename="work_dir/crops/coco_minitrain_25kimagestrain2017000000320957.jpg_183_300_10_10.jpg")

Out[14]:

The following image is labeled as a person in the dataset.

In [15]:

Image(filename="work_dir/crops/coco_minitrain_25kimagestrain2017000000407413.jpg_301_175_10_10.jpg")

Out[15]:

Considering the image size, it is difficult to discern if the label is correct.

Let's also investigate objects with extreme aspect ratios.

In [16]:

aspect_ratio_objects.head(3)

Out[16]:

	col_x	row_y	width	height	label	split	crop_filename	filename	index	is_valid	area	aspect
134879	230.69	1.11	16.64	419.24	kite	train	work_dir/crops/coco_minitrain_25kimagestrain2017000000093298.jpg_230_1_16_419.jpg	coco_minitrain_25k/images/train2017/000000093298.jpg	121215.0	True	6976.1536	0.039691
3644	1.92	136.50	11.51	263.87	person	train	work_dir/crops/coco_minitrain_25kimagestrain2017000000002444.jpg_1_136_11_263.jpg	coco_minitrain_25k/images/train2017/000000002444.jpg	3259.0	True	3037.1437	0.043620
164234	222.84	63.49	16.19	318.20	spoon	train	work_dir/crops/coco_minitrain_25kimagestrain2017000000116502.jpg_222_63_16_318.jpg	coco_minitrain_25k/images/train2017/000000116502.jpg	147609.0	True	5151.6580	0.050880

The following object is labeled kite in the dataset.

In [17]:

Image(filename="work_dir/crops/coco_minitrain_25kimagestrain2017000000093298.jpg_230_1_16_419.jpg")

Out[17]:

In [18]:

aspect_ratio_objects.tail(3)

Out[18]:

	col_x	row_y	width	height	label	split	crop_filename	filename	index	is_valid	area	aspect
4263	33.00	216.00	602.00	18.00	boat	train	work_dir/crops/coco_minitrain_25kimagestrain2017000000527098.jpg_33_216_602_18.jpg	coco_minitrain_25k/images/train2017/000000527098.jpg	3827.0	True	10836.0000	33.444444
77862	1.12	58.60	638.88	15.74	bench	train	work_dir/crops/coco_minitrain_25kimagestrain2017000000444692.jpg_1_58_638_15.jpg	coco_minitrain_25k/images/train2017/000000444692.jpg	70136.0	True	10055.9712	40.589581
173236	17.27	197.91	601.64	13.43	train	train	work_dir/crops/coco_minitrain_25kimagestrain2017000000516740.jpg_17_197_601_13.jpg	coco_minitrain_25k/images/train2017/000000516740.jpg	155696.0	True	8080.0252	44.798213

The following object is labeled boat in the dataset.

In [19]:

Image(filename="work_dir/crops/coco_minitrain_25kimagestrain2017000000527098.jpg_33_216_602_18.jpg")

Out[19]:

Using fastdup, we have discovered many items that are either tiny (10x10 pixels) or have extreme aspect ratios!

Bad Bounding Boxes¶

Bounding boxes that are either too small or go beyond image boundaries are flagged as a bad bounding box in fastdup.

We can get a list of bad bounding boxes by reading the atrain_features.bad.csv file.

In [20]:

bad_bb = pd.read_csv('work_dir/full_image_run/atrain_features.bad.csv')

In [21]:

bad_bb

Out[21]:

	index	filename	error_code
0	16	coco_minitrain_25k/images/train2017/000000262162.jpg	ERROR_BAD_BOUNDING_BOX
1	60	coco_minitrain_25k/images/train2017/000000524325.jpg	ERROR_BAD_BOUNDING_BOX
2	65	coco_minitrain_25k/images/train2017/000000524325.jpg	ERROR_BAD_BOUNDING_BOX
3	66	coco_minitrain_25k/images/train2017/000000524325.jpg	ERROR_BAD_BOUNDING_BOX
4	89	coco_minitrain_25k/images/train2017/000000131152.jpg	ERROR_BAD_BOUNDING_BOX
...	...	...	...
18587	183487	coco_minitrain_25k/images/train2017/000000131007.jpg	ERROR_BAD_BOUNDING_BOX
18588	183494	coco_minitrain_25k/images/train2017/000000131011.jpg	ERROR_BAD_BOUNDING_BOX
18589	183504	coco_minitrain_25k/images/train2017/000000262086.jpg	ERROR_BAD_BOUNDING_BOX
18590	183505	coco_minitrain_25k/images/train2017/000000262086.jpg	ERROR_BAD_BOUNDING_BOX
18591	183530	coco_minitrain_25k/images/train2017/000000262093.jpg	ERROR_BAD_BOUNDING_BOX

18592 rows × 3 columns

In [22]:

bad_bb['error_code'].value_counts()

Out[22]:

error_code
ERROR_BAD_BOUNDING_BOX    18592
Name: count, dtype: int64

We have found 18,592 (!) bounding boxes that are either too small or go beyond image boundaries. This is 10% of the data! Filtering them out would both save us gruesome debugging of training errors or failures, and help up provide the model with useful size objects.

Possible Mislabels¶

The fastdup similarity search and similarity gallery are strong tools for finding objects that are possibly mislabeled. By finding each object's nearest neighbors and their classes, we can find objects with classes contradicting their neighbors' (a strong sign of mislabels).

Running the similarity gallery shows if an image has high similarity with two of its closest neighbors, yet has different labels. This helps surface potential mislabeling in the dataset.

In [23]:

fd.vis.similarity_gallery(slice='diff')

Warning: you are running create_similarity_gallery() without providing get_label_func so similarities are not computed between different classes. It is recommended to run this report with labels. Without labels this report output is similar to create_duplicate_gallery()

100%|█████████████████████████████████████████████████████████████████████████████████████████████████████████████████████| 45406/45406 [00:01<00:00, 24065.76it/s]
100%|█████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████| 20/20 [00:00<00:00, 291.38it/s]

Stored similar images visual view in  work_dir/galleries/similarity.html
########################################################################################
Would you like to see awesome visualizations for some of the most popular academic datasets?
Click here to see and learn more: https://app.visual-layer.com/vl-datasets?utm_source=fastdup
########################################################################################

Similarity Report, label_score

Info From
label	dog
from	/crops/imagestrain2017000000000109.jpg_539_295_22_18.jpg

Info To
0.919005	/crops/imagestrain2017000000250069.jpg_323_234_22_15.jpg	cell phone
0.914812	/crops/imagestrain2017000000535160.jpg_174_317_18_17.jpg	bird

Query Image

Similar

Info From
label	spoon
from	/crops/imagestrain2017000000000294.jpg_388_387_14_22.jpg

Info To
0.969069	/crops/imagestrain2017000000033687.jpg_370_252_18_23.jpg	car
0.968792	/crops/imagestrain2017000000292510.jpg_432_178_14_16.jpg	sports ball

Query Image

Similar

Info From
label	spoon
from	/crops/imagestrain2017000000000294.jpg_392_299_16_32.jpg

Info To
0.949718	/crops/imagestrain2017000000005505.jpg_596_392_16_43.jpg	handbag
0.948114	/crops/imagestrain2017000000443641.jpg_619_409_20_27.jpg	person

Query Image

Similar

Info From
label	spoon
from	/crops/imagestrain2017000000000294.jpg_401_323_23_25.jpg

Info To
0.909022	/crops/imagestrain2017000000131374.jpg_254_107_16_25.jpg	person
0.902268	/crops/imagestrain2017000000255013.jpg_162_213_23_23.jpg	person

Query Image

Similar

Info From
label	book
from	/crops/imagestrain2017000000000384.jpg_160_509_32_16.jpg

Info To
0.920194	/crops/imagestrain2017000000547528.jpg_158_473_31_17.jpg	car
0.909746	/crops/imagestrain2017000000211963.jpg_4_218_29_19.jpg	car

Query Image

Similar

Info From
label	car
from	/crops/imagestrain2017000000000540.jpg_162_394_15_10.jpg

Info To
0.924145	/crops/imagestrain2017000000179926.jpg_399_468_14_11.jpg	person
0.920198	/crops/imagestrain2017000000382924.jpg_198_154_13_10.jpg	cup

Query Image

Similar

Info From
label	truck
from	/crops/imagestrain2017000000000540.jpg_329_192_11_13.jpg

Info To
0.978024	/crops/imagestrain2017000000161738.jpg_270_209_15_13.jpg	car
0.977553	/crops/imagestrain2017000000182104.jpg_90_210_11_14.jpg	car

Query Image

Similar

Info From
label	bus
from	/crops/imagestrain2017000000000540.jpg_416_203_23_10.jpg

Info To
0.950101	/crops/imagestrain2017000000117721.jpg_574_313_42_10.jpg	chair
0.947586	/crops/imagestrain2017000000120099.jpg_457_116_22_10.jpg	book

Query Image

Similar

Info From
label	bus
from	/crops/imagestrain2017000000000540.jpg_452_385_13_19.jpg

Info To
0.967401	/crops/imagestrain2017000000508488.jpg_161_265_14_21.jpg	car
0.962299	/crops/imagestrain2017000000540554.jpg_111_221_13_19.jpg	bench

Query Image

Similar

Info From
label	truck
from	/crops/imagestrain2017000000000540.jpg_93_392_12_10.jpg

Info To
0.988809	/crops/imagestrain2017000000000540.jpg_94_391_12_10.jpg	car
0.978279	/crops/imagestrain2017000000021083.jpg_371_307_12_10.jpg	car

Query Image

Similar

Info From
label	clock
from	/crops/imagestrain2017000000000801.jpg_376_90_12_51.jpg

Info To
0.933241	/crops/imagestrain2017000000139344.jpg_489_273_12_69.jpg	giraffe
0.928798	/crops/imagestrain2017000000051465.jpg_67_335_12_41.jpg	person

Query Image

Similar

Info From
label	person
from	/crops/imagestrain2017000000000984.jpg_2_135_469_500.jpg

Info To
0.971609	/crops/imagestrain2017000000000984.jpg_145_6_334_606.jpg	orange
0.910559	/crops/imagestrain2017000000000984.jpg_85_254_229_305.jpg	banana

Query Image

Similar

Info From
label	baseball glove
from	/crops/imagestrain2017000000001270.jpg_181_250_19_30.jpg

Info To
0.952877	/crops/imagestrain2017000000391351.jpg_350_329_16_32.jpg	person
0.952217	/crops/imagestrain2017000000511814.jpg_62_197_17_29.jpg	motorcycle

Query Image

Similar

Info From
label	person
from	/crops/imagestrain2017000000001350.jpg_4_274_10_44.jpg

Info To
0.947479	/crops/imagestrain2017000000522947.jpg_285_280_10_27.jpg	wine glass
0.918168	/crops/imagestrain2017000000501624.jpg_324_239_10_44.jpg	tie

Query Image

Similar

Info From
label	traffic light
from	/crops/imagestrain2017000000001523.jpg_434_117_10_52.jpg

Info To
0.93222	/crops/imagestrain2017000000258095.jpg_405_120_10_33.jpg	person
0.928663	/crops/imagestrain2017000000477636.jpg_158_237_10_36.jpg	person

Query Image

Similar

Info From
label	knife
from	/crops/imagestrain2017000000001573.jpg_350_8_21_24.jpg

Info To
0.937601	/crops/imagestrain2017000000409771.jpg_442_239_28_19.jpg	car
0.925864	/crops/imagestrain2017000000547999.jpg_117_208_19_23.jpg	remote

Query Image

Similar

Info From
label	spoon
from	/crops/imagestrain2017000000001573.jpg_385_19_10_24.jpg

Info To
0.977106	/crops/imagestrain2017000000215616.jpg_538_230_10_19.jpg	person
0.976037	/crops/imagestrain2017000000093894.jpg_184_215_10_25.jpg	bottle

Query Image

Similar

Info From
label	person
from	/crops/imagestrain2017000000001580.jpg_247_174_14_15.jpg

Info To
0.975657	/crops/imagestrain2017000000150989.jpg_224_168_15_20.jpg	dog
0.974637	/crops/imagestrain2017000000577924.jpg_211_243_12_16.jpg	handbag

Query Image

Similar

Info From
label	car
from	/crops/imagestrain2017000000001720.jpg_0_241_16_19.jpg

Info To
0.953981	/crops/imagestrain2017000000298269.jpg_206_268_11_15.jpg	person
0.953843	/crops/imagestrain2017000000365822.jpg_275_269_11_36.jpg	bottle

Query Image

Similar

Info From
label	person
from	/crops/imagestrain2017000000001720.jpg_542_274_15_16.jpg

Info To
0.951418	/crops/imagestrain2017000000065948.jpg_475_266_11_24.jpg	wine glass
0.95006	/crops/imagestrain2017000000385098.jpg_294_99_16_24.jpg	handbag

Query Image

Similar

Out[23]:

	from	to	label	label2	distance	score	length
6	work_dir/crops/coco_minitrain_25kimagestrain2017000000000109.jpg_539_295_22_18.jpg	[work_dir/crops/coco_minitrain_25kimagestrain2017000000535160.jpg_174_317_18_17.jpg, work_dir/crops/coco_minitrain_25kimagestrain2017000000250069.jpg_323_234_22_15.jpg]	[dog, dog]	[bird, cell phone]	[0.914812, 0.919005]	0.0	2
12	work_dir/crops/coco_minitrain_25kimagestrain2017000000000294.jpg_388_387_14_22.jpg	[work_dir/crops/coco_minitrain_25kimagestrain2017000000292510.jpg_432_178_14_16.jpg, work_dir/crops/coco_minitrain_25kimagestrain2017000000033687.jpg_370_252_18_23.jpg]	[spoon, spoon]	[sports ball, car]	[0.968792, 0.969069]	0.0	2
13	work_dir/crops/coco_minitrain_25kimagestrain2017000000000294.jpg_392_299_16_32.jpg	[work_dir/crops/coco_minitrain_25kimagestrain2017000000443641.jpg_619_409_20_27.jpg, work_dir/crops/coco_minitrain_25kimagestrain2017000000005505.jpg_596_392_16_43.jpg]	[spoon, spoon]	[person, handbag]	[0.948114, 0.949718]	0.0	2
14	work_dir/crops/coco_minitrain_25kimagestrain2017000000000294.jpg_401_323_23_25.jpg	[work_dir/crops/coco_minitrain_25kimagestrain2017000000255013.jpg_162_213_23_23.jpg, work_dir/crops/coco_minitrain_25kimagestrain2017000000131374.jpg_254_107_16_25.jpg]	[spoon, spoon]	[person, person]	[0.902268, 0.909022]	0.0	2
28	work_dir/crops/coco_minitrain_25kimagestrain2017000000000384.jpg_160_509_32_16.jpg	[work_dir/crops/coco_minitrain_25kimagestrain2017000000211963.jpg_4_218_29_19.jpg, work_dir/crops/coco_minitrain_25kimagestrain2017000000547528.jpg_158_473_31_17.jpg]	[book, book]	[car, car]	[0.909746, 0.920194]	0.0	2
...	...	...	...	...	...	...	...
45398	work_dir/crops/coco_minitrain_25kimagestrain2017000000581900.jpg_282_147_10_23.jpg	[work_dir/crops/coco_minitrain_25kimagestrain2017000000043286.jpg_429_244_10_31.jpg, work_dir/crops/coco_minitrain_25kimagestrain2017000000017057.jpg_572_197_10_37.jpg]	[person, person]	[person, person]	[0.947242, 0.947285]	100.0	2
45399	work_dir/crops/coco_minitrain_25kimagestrain2017000000581900.jpg_290_150_15_21.jpg	[work_dir/crops/coco_minitrain_25kimagestrain2017000000342532.jpg_426_0_15_15.jpg, work_dir/crops/coco_minitrain_25kimagestrain2017000000497617.jpg_254_264_14_18.jpg]	[person, person]	[person, person]	[0.952342, 0.953701]	100.0	2
45400	work_dir/crops/coco_minitrain_25kimagestrain2017000000581900.jpg_305_153_15_39.jpg	[work_dir/crops/coco_minitrain_25kimagestrain2017000000490347.jpg_475_125_15_32.jpg, work_dir/crops/coco_minitrain_25kimagestrain2017000000213312.jpg_354_14_18_30.jpg]	[person, person]	[person, person]	[0.940181, 0.943313]	100.0	2
3214	work_dir/crops/coco_minitrain_25kimagestrain2017000000041773.jpg_466_104_70_63.jpg	[work_dir/crops/coco_minitrain_25kimagestrain2017000000041773.jpg_466_104_70_63.jpg, work_dir/crops/coco_minitrain_25kimagestrain2017000000041773.jpg_466_104_70_63.jpg, work_dir/crops/coco_minitrain_25kimagestrain2017000000041773.jpg_466_104_70_63.jpg, work_dir/crops/coco_minitrain_25kimagestrain2017000000041773.jpg_466_104_70_63.jpg, work_dir/crops/coco_minitrain_25kimagestrain2017000000041773.jpg_466_104_70_63.jpg, work_dir/crops/coco_minitrain_25kimagestrain2017000000041773.jpg_466_104_70_63.jpg, work_dir/crops/coco_minitrain_25kimagestrain2017000000041773.jpg_466_104_70_63.jpg, work_dir/crops/coco_minitrain_25kimagestrain2017000000041773.jpg_466_104_70_63.jpg]	[bird, bird, bird, bird, bird, bird, bird, bird]	[bird, bird, bird, bird, bird, bird, bird, bird]	[1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0]	100.0	8
39587	work_dir/crops/coco_minitrain_25kimagestrain2017000000508538.jpg_386_137_25_19.jpg	[work_dir/crops/coco_minitrain_25kimagestrain2017000000508538.jpg_386_137_25_19.jpg, work_dir/crops/coco_minitrain_25kimagestrain2017000000508538.jpg_386_137_25_19.jpg, work_dir/crops/coco_minitrain_25kimagestrain2017000000508538.jpg_386_137_25_19.jpg, work_dir/crops/coco_minitrain_25kimagestrain2017000000508538.jpg_386_137_25_19.jpg, work_dir/crops/coco_minitrain_25kimagestrain2017000000508538.jpg_386_137_25_19.jpg, work_dir/crops/coco_minitrain_25kimagestrain2017000000508538.jpg_386_137_25_19.jpg, work_dir/crops/coco_minitrain_25kimagestrain2017000000508538.jpg_386_137_25_19.jpg, work_dir/crops/coco_minitrain_25kimagestrain2017000000508538.jpg_386_137_25_19.jpg]	[potted plant, potted plant, potted plant, potted plant, potted plant, potted plant, potted plant, potted plant]	[potted plant, potted plant, potted plant, potted plant, potted plant, potted plant, potted plant, potted plant]	[1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0]	100.0	8

30248 rows × 7 columns

Wrap Up¶

Next, feel free to check out other tutorials -

⚡ Quickstart: Learn how to install fastdup, load a dataset and analyze it for potential issues such as duplicates/near-duplicates, broken images, outliers, dark/bright/blurry images, and view visually similar image clusters. If you're new, start here!
🧹 Clean Image Folder: Learn how to analyze and clean a folder of images from potential issues and export a list of problematic files for further action. If you have an unorganized folder of images, this is a good place to start.
🖼 Analyze Image Classification Dataset: Learn how to load a labeled image classification dataset and analyze for potential issues. If you have labeled ImageNet-style folder structure, have a go!
🎁 Analyze Object Detection Dataset: Learn how to load bounding box annotations for object detection and analyze for potential issues. If you have a COCO-style labeled object detection dataset, give this example a try.

VL Profiler¶

If you prefer a no-code platform to inspect and visualize your dataset, try our free cloud product VL Profiler - VL Profiler is our first no-code commercial product that lets you visualize and inspect your dataset in your browser.

Sign up now, it's free.

As usual, feedback is welcome!

Questions? Drop by our Slack channel or open an issue on GitHub.