File size: 6,760 Bytes
fe1345c
0b21e1c
fe1345c
0b21e1c
 
 
 
 
 
 
 
 
 
 
 
fe1345c
b3ea911
0b21e1c
 
 
 
8745cf9
 
 
0b21e1c
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
717573e
 
0b21e1c
 
 
 
 
 
 
0f486fd
0b21e1c
b2bb2bb
 
 
 
 
 
0b21e1c
 
 
 
 
9945ae3
8745cf9
 
 
 
 
 
 
 
 
e5afa48
8745cf9
 
 
 
 
 
 
 
 
 
0b21e1c
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
d2925c6
 
b2bb2bb
 
9945ae3
d2925c6
b2bb2bb
 
d2925c6
 
b2bb2bb
 
d2925c6
 
0b21e1c
 
d2925c6
0b21e1c
 
 
 
 
9945ae3
0b21e1c
 
 
b2bb2bb
0b21e1c
 
 
d2925c6
0b21e1c
 
 
 
 
d2925c6
0b21e1c
 
 
e8fdce3
0b21e1c
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
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
---
pretty_name: Ecoset
license: cc
source_datasets:
- original
task_categories:
- image-classification
- image
task_ids:
- multi-class-image-classification
- other-other-image-classification
- image-classification
- other-image-classification
paperswithcode_id: ecoset

---

## Table of Contents
- [Dataset Description](#dataset-description)
  - [Dataset Summary](#dataset-summary)
  - [Supported Tasks and Leaderboards](#supported-tasks-and-leaderboards)
- [Installation](#installation)
  - [Install requirements](#install-requirements)
  - [Download settings](#download-settings)
- [Dataset Structure](#dataset-structure)
  - [Data Instances](#data-instances)
  - [Data Fields](#data-fields)
  - [Data Splits](#data-splits)
- [Dataset Creation](#dataset-creation)
  - [Personal and Sensitive Information](#personal-and-sensitive-information)
- [Considerations for Using the Data](#considerations-for-using-the-data)
  - [Social Impact of Dataset](#social-impact-of-dataset)
  - [Discussion of Biases](#discussion-of-biases)
  - [Other Known Limitations](#other-known-limitations)
- [Additional Information](#additional-information)
  - [Dataset Curators](#dataset-curators)
  - [Licensing Information](#licensing-information)
  - [Citation Information](#citation-information)
  - [Contributions](#contributions)
  
  
## Dataset Description

- **Homepage:** [https://www.kietzmannlab.org/ecoset](https://www.kietzmannlab.org/ecoset/)
- **Repository:** [https://codeocean.com/capsule/9570390/tree/v1](https://codeocean.com/capsule/9570390/tree/v1)
- **Paper:** [https://www.pnas.org/doi/full/10.1073/pnas.2011417118](https://doi.org/10.1073/pnas.2011417118)
- **Point of Contact:** [tim.kietzmann@uni-osnabrueck.de](tim.kietzmann@uni-osnabrueck.de)


### Dataset Summary

Tired of all the dogs in ImageNet (ILSVRC)? Then ecoset is here for you. 1.5m images 
from 565 basic level categories, chosen to be both (i) frequent in linguistic usage, 
and (ii) rated by human observers as concrete (e.g. ‘table’ is concrete, ‘romance’ 
is not).

Ecoset is a typical image recognition dataset, combining images of objects with appropriate 
labels (one label per image). Importantly, ecoset is intended to provide higher ecological 
validity than its counterparts, with a mislabelling error rate < 5% and filtered for NSFW content.
For more information on the dataset, consider reading the [original publication](https://doi.org/10.1073/pnas.2011417118).

Ecoset consists of a train, test, and validation subset which all are openly available to the user.

### Supported Tasks and Leaderboards

Ecoset is a large multi-class single-label object recognition image dataset (similar to ImageNet).

## Installation

### Install Requirements

In order to work with ecoset, please make sure to install the s3 compatible version of huggingface datasets, which should include the `s3fs`, `botocore` and `boto3` modules:

```bash
pip install datasets[s3]
```


### Download Settings

Please set `ignore_verifications=True`. when downloading this dataset, else the download will result in an error:

```python
from datasets import load_dataset

dataset = load_dataset("DiGyt/ecoset", ignore_verifications=True)
```


## Dataset Structure


We show detailed information for all the configurations of the dataset. Currently, there is only one setting (`Full`) available, containing all data.


### Data Instances

#### Full

- **Size of downloaded dataset files:** 155 GB
- **Total amount of disk used:** 311 GB

## Dataset Creation

A total of 565 categories were selected based on the following: 1) their word frequency in American television and film subtitles (SUBTLEX_US), 2) the perceived concreteness by human observers, and 3) the availability of a minimum of 700 images. Images were sourced via the overall ImageNet database (the same resource used for ILSVRC 2012) or obtained under CC BY-NC-SA 2.0 license from Bing image search and Flickr. Thorough data cleaning procedures were put in place to remove duplicates and to assure an expected misclassification rate per category of <4%.

### Curation Rationale

...TODO

### Source Data

The source data is available under: [https://codeocean.com/capsule/9570390/tree/v1](https://codeocean.com/capsule/9570390/tree/v1)

### Annotations

Each ecoset image folder is annotated with class labels according to the main object depicted in a class of images. No further annotations are added to the dataset.

### Personal and Sensitive Information

The dataset was tested to exclude sensitive images using Yahoo's Open NSFW detection model, removing all image with an NSFW score above 0.8. For this dataset, only images with secured license information was used, which should prevent the inclusion of images without consent of the image's authors and subjects. Despite these measures, it is possible that the images in the dataset contain personal and sensitive information.
  
## Considerations for Using the Data

### Social Impact of Dataset
  
...TODO

### Discussion of Biases

Despite best efforts to provide an ecologically valid dataset, ecoset is likely to contain biased data. The category selection of ecoset was based on human concreteness ratings and word frequencies in a corpus consisting of American television and film subtitles. This undoubtedly biases the category selection toward Western cultures. Image inclusion was based on the availability via Bing/Flickr search results as well as the existence of relevant ImageNet categories. Images depicting people, specifically the categories “man,” “woman,” and “child,” were not sampled according to census distributions (age, ethnicity, gender, etc.).

### Other Known Limitations

In addition to points mentioned in [Discussion of Biases](#discussion-of-biases), ecoset image and category distributions do not reflect the naturalistic, egocentric visual input typically encountered in the everyday life of infant and adults.

## Additional Information

### Dataset Curators

The corpus was put together by Johannes Mehrer, Courtney J. Spoerer, Emer C. Jones, Nikolaus Kriegeskorte, and Tim C. Kietzmann.

### Licensing Information

Ecoset is licensed under Creative Commons Attribution-NonCommercial-ShareAlike 2.0 license (cc-by-nc-sa-2.0).

### Citation Information

```
@article{mehrer2021ecologically,
  title={An ecologically motivated image dataset for deep learning yields better models of human vision},
  author={Mehrer, Johannes and Spoerer, Courtney J and Jones, Emer C and Kriegeskorte, Nikolaus and Kietzmann, Tim C},
  journal={Proceedings of the National Academy of Sciences},
  volume={118},
  number={8},
  pages={e2011417118},
  year={2021},
  publisher={National Acad Sciences}
}
```

### Contributions

Thanks to  #TODO