Datasets:

Oztobuzz
/

ImageNet_10k

idx int64 0 10k	label stringclasses 998 values	gt_ans stringclasses 2 values
0	coyote	yes
1	ostrich	no
2	four-poster	yes
3	wallet	no
4	measuring_cup	yes
5	ocarina	no
6	ski	yes
7	Polaroid_camera	no
8	snowplow	yes
9	American_Staffordshire_terrier	no
10	pickup	yes
11	cardoon	no
12	Shetland_sheepdog	yes
13	gas_pump	no
14	beaker	yes
15	broccoli	no
16	bookcase	yes
17	lacewing	no
18	English_setter	yes
19	quill	no
20	dogsled	yes
21	bathing_cap	no
22	dingo	yes
23	red-breasted_merganser	no
24	Doberman	yes
25	sturgeon	no
26	perfume	yes
27	German_shepherd	no
28	langur	yes
29	banjo	no
30	wok	yes
31	miniature_pinscher	no
32	solar_dish	yes
33	sarong	no
34	cheetah	yes
35	jeep	no
36	tusker	yes
37	cassette_player	no
38	wire-haired_fox_terrier	yes
39	doormat	no
40	grey_whale	yes
41	gazelle	no
42	thunder_snake	yes
43	komondor	no
44	Welsh_springer_spaniel	yes
45	academic_gown	no
46	vacuum	yes
47	dogsled	no
48	home_theater	yes
49	mitten	no
50	perfume	yes
51	grasshopper	no
52	briard	yes
53	kimono	no
54	printer	yes
55	Model_T	no
56	harmonica	yes
57	drum	no
58	washer	yes
59	ice_lolly	no
60	weasel	yes
61	goldfish	no
62	broom	yes
63	ambulance	no
64	pickelhaube	yes
65	spotted_salamander	no
66	dhole	yes
67	Arabian_camel	no
68	mink	yes
69	bolo_tie	no
70	nematode	yes
71	coffeepot	no
72	mailbag	yes
73	handkerchief	no
74	weevil	yes
75	sundial	no
76	minivan	yes
77	carousel	no
78	macaque	yes
79	tractor	no
80	grand_piano	yes
81	carpenter's_kit	no
82	slide_rule	yes
83	bicycle-built-for-two	no
84	ski	yes
85	binoculars	no
86	boxer	yes
87	church	no
88	thresher	yes
89	grocery_store	no
90	thimble	yes
91	brass	no
92	basketball	yes
93	water_ouzel	no
94	water_bottle	yes
95	scuba_diver	no
96	accordion	yes
97	lipstick	no
98	wood_rabbit	yes
99	Madagascar_cat	no

End of preview. Expand in Data Studio

Dataset Card for ImageNet_10k Dataset

This dataset is derived from ImageNet and contains 10,000 image-label pairs, designed for binary classification in object detection tasks.

Dataset Details

Dataset Description

This dataset consists of 10,000 image-label pairs sampled from ImageNet. 5,000 pairs have correct image-label matches (positive examples labeled "yes"), and 5,000 pairs have random labels assigned from the ImageNet 1000-class taxonomy (negative examples labeled "no"). The dataset is intended for training and evaluating object detection systems with a binary classification component.

Language(s) (NLP): English (for labels)
License: The dataset inherits the ImageNet license terms (Custom license - requires acceptance of Terms of Access)

Dataset Sources

Repository: Based on ImageNet (https://image-net.org/)
Paper: Based on ImageNet (Deng, J., Dong, W., Socher, R., Li, L.-J., Li, K., & Fei-Fei, L. (2009). ImageNet: A Large-Scale Hierarchical Image Database. CVPR 2009.)

Uses

Direct Use

This dataset is suitable for:

Training binary object detection classifiers
Evaluating object recognition systems' accuracy
Testing models' ability to verify if an image contains a specified object
Benchmarking computer vision systems on object verification tasks

Dataset Structure

The dataset consists of 10,000 examples with the following fields:

idx: int64 - A unique identifier for each example
image: image - The image data from ImageNet
label: string - The object label being verified (from ImageNet's 1000 classes)
gt_ans: string - Ground truth answer ("yes" for correct label, "no" for random incorrect label)

The dataset is evenly split between positive examples (5,000 images with their correct ImageNet labels, marked "yes") and negative examples (5,000 images with randomly assigned incorrect labels from ImageNet's taxonomy, marked "no").

Dataset Creation

Curation Rationale

This dataset was created to provide a balanced binary classification task for object detection systems. By including both correct and incorrect image-label pairs, it supports the development of models that can verify whether a specific object appears in an image.

Source Data

Data Collection and Processing

5,000 images were randomly sampled from ImageNet along with their correct labels
5,000 additional images were sampled and assigned random labels from the ImageNet 1000-class taxonomy
All correct label pairs were marked with "yes" in the gt_ans field
All random label pairs were marked with "no" in the gt_ans field
The complete set of 10,000 examples was assigned unique indices

Who are the source data producers?

The original images come from ImageNet, which collected images from the web. The labels were originally created by ImageNet annotators through a combination of automated and manual processes. The binary classification labels ("yes"/"no") were added during the creation of this derivative dataset.

Personal and Sensitive Information

This dataset inherits the privacy considerations of ImageNet. While efforts were made in ImageNet to remove certain personally identifiable information, users should be aware that the images may contain people, locations, or other potentially identifying information. No additional personal data was introduced during the creation of this derivative dataset.

Recommendations

Users should:

Be aware of ImageNet's documented biases when using this dataset
Evaluate model performance across different object categories to identify potential performance disparities
Consider augmenting with more diverse data sources for production applications
Use this as a benchmark or starting point rather than a complete solution for production object detection

APA: ImageNet Object Detection Dataset. (2025). Derived from ImageNet by Deng et al., 2009.

Dataset Card Contact

[Your contact information here]

Downloads last month: 78

Size of downloaded dataset files:

952 MB

Size of the auto-converted Parquet files:

952 MB

Number of rows:

10,000