This dataset is a subset of ObjectNet with 104 classes overlapping with ImageNet's classes. This dataset is originally proposed in OODRobustBench to simulate a type of natural distribution shift to evaluate Out-Of-Distribution (OOD) performance for ImageNet-trained models. This subset and the full ObjectNet are provided under a license derived from Creative Commons Attribution 4.0 with only two additional clauses.

1. ObjectNet may never be used to tune the parameters of any model.
2. Any individual images from ObjectNet may only be posted to the web including their 1 pixel red border.

Note that as required by the above license the images in this dataset contain 1 pixel red border.

We sincerely appreciate the creators of ObjectNet for collecting and curating ObjectNet.

NOTE: BY USING THIS DATASET YOU ACKNOWLEDGE THAT YOU HAVE READ THE LICENSE AND WILL ABIDE BY THE TERMS THEREWITHIN

THE LICENSE

ObjectNet is free to use for both research and commercial
applications. The authors own the source images and allow their use
under a license derived from Creative Commons Attribution 4.0 with
two additional clauses:

1. ObjectNet may never be used to tune the parameters of any
  model. This includes, but is not limited to, computing statistics
  on ObjectNet and including those statistics into a model,
  fine-tuning on ObjectNet, performing gradient updates on any
  parameters based on these images.

2. Any individual images from ObjectNet may only be posted to the web
  including their 1 pixel red border.

If you post this archive in a public location, please leave the password
intact as "objectnetisatestset".

[Other General License Information Conforms to Attribution 4.0 International]

This is Part 2 of 10 * Original Paper Link * ObjectNet Website

The links to the various parts of the dataset are:

• Part 1
• Part 2
• Part 3
• Part 4
• Part 5
• Part 6
• Part 7
• Part 8
• Part 9
• Part 10

Description From ObjectNET Homepage

What is ObjectNet?

• A new kind of vision dataset borrowing the idea of controls from other areas of science.
• No training set, only a test set! Put your vision system through its paces.
• Collected to intentionally show objects from new viewpoints on new backgrounds.
• 50,000 image test set, same as ImageNet, with controls for rotation, background, and viewpoint.
• 313 object classes with 113 overlapping ImageNet
• Large performance drop, what you can expect from vision systems in the real world!
• Robust to fine-tuning and a very difficult transfer learning problem

Controls For Biases Increase Variation

https://objectnet.dev/images/objectnet_controls_table.png">

Easy For Humans, Hard For Machines

• Ready to help develop the next generation of object recognition algorithms that have robustness, bias, and safety in mind.
• Controls can remove bias from other datasets machine learning, not just vision.

https://objectnet.dev/images/objectnet_results.png">

Full Description

ObjectNet is a large real-world test set for object recognition with control where object backgrounds, rotations, and imaging viewpoints are random.

Most scientific experiments have controls, confounds which are removed from the data, to ensure that subjects cannot perform a task by exploiting trivial correlations in the data. Historically, large machine learning and computer vision datasets have lacked such controls. This has resulted in models that must be fine-tuned for new datasets and perform better on datasets than in real-world applications. When tested on ObjectNet, object detectors show a 40-45% drop in performance, with respect to their performance on other benchmarks, due to the controls for biases. Controls make ObjectNet robust to fine-tuning showing only small performance increases.

We develop a highly automated platform that enables gathering datasets with controls by crowdsourcing image capturing and annotation. ObjectNet is the same size as the ImageNet test set (50,000 images), and by design does not come paired with a training set in order to encourage generalization. The dataset is both easier than ImageNet – objects are largely centred and unoccluded – and harder, due to the controls. Although we focus on object recognition here, data with controls can be gathered at scale using automated tools throughout machine learning to generate datasets that exercise models in new ways thus providing valuable feedback to researchers. This work opens up new avenues for research in generalizable, robust, and more human-like computer vision and in creating datasets where results are predictive of real-world performance.

Citation

...

ObjectNet

A webp (lossless) encoded version of ObjectNet-1.0 at original resolution.

  License / Usage Terms

ObjectNet is free to use for both research and commercial applications. The authors own the source images and allow their use under a license derived from Creative Commons Attribution 4.0 with only two additional clauses.

ObjectNet may never be used to tune the parameters of any model. Any individual images from ObjectNet may only be posted to the web including… See the full description on the dataset page: https://huggingface.co/datasets/timm/objectnet.

"ObjectNet is a test set of images collected directly using crowd-sourcing. A new kind of vision dataset borrowing the idea of controls from other areas of science. No training set, only a test set! Put your vision system through its paces. Collected to intentionally show objects from new viewpoints on new backgrounds. 50,000 image test set, same as ImageNet, with controls for rotation, background, and viewpoint. 313 object classes with 113 overlapping ImageNet Large performance drop, what you can expect from vision systems in the real world! Robust to fine-tuning and a very difficult transfer learning problem"

ObjectNet (720P Shortest Edge, ImageNet-1k Overlap)

A webp (lossless) encoded version of ObjectNet-1.0 resized to shortest edge = 720 pixels. Containing only the 113 classes that overlap with ImageNet-1k.

  License / Usage Terms

ObjectNet is free to use for both research and commercial applications. The authors own the source images and allow their use under a license derived from Creative Commons Attribution 4.0 with only two additional clauses.

ObjectNet may never be… See the full description on the dataset page: https://huggingface.co/datasets/timm/objectnet-720p-in1k.