This release is a pre-release and may not be stable for production use.
Welcome! This is the official code for the ICML 2022 workshop on crowdsourcing novel metrics and test datasets for model evaluation on ImageNet scale.
While the popularity of robustness benchmarks and new test datasets increased over the past years, the performance of computer vision models is still largely evaluated on ImageNet directly, or on simulated or isolated distribution shifts like in ImageNet-C.
Goal: This workshop aims to enhance and consolidate the landscape of robustness evaluation datasets for computer vision and collect new test sets and metrics for quantifying desirable or problematic properties of computer vision models. Our goal is to bring the robustness, domain adaptation, and out-of-distribution detection communities together to work on a new broad-scale benchmark that tests diverse aspects of current computer vision models and guides the way towards the next generation of models.
Over the course of the workshop, the package will be populated with new tasks, metrics and datasets that are suitable for understanding the properties of ImageNet scaled computer vision models beyond the metrics that are typically reported.
How to use this benchmark
For now, you can use the benchmark by installing it from this repository:
$ pip install git+https://github.com/shift-happens-benchmark/icml-2022.gitAfter the workshop you will be able to use this package and all its included tasks with:
$ pip install shifthappensHow to contribute
Since the aim of this workshop & package is to build a unified platform for datasets investigating and highlighting interesting properties of ImageNet-scale vision models, we are looking for your contribution. If you decide to contribute a new task to the benchmark before ICML 2022 please consider officially submitting it to the workshop - for more details see here.
Adding tasks and datasets
Tasks in this benchmark package should highlight interesting properties of vision models. For one, this means that you can integrate new datasets you built. In addition, you can also propose new evaluation schemes (i.e. new tasks) for already existing datasets, like test-time adaptation evaluation on robustness datasets. You can think about examples/scenarios that might be of interest for industrial applications just as well as purely academic examples - as long as the new tasks/datasets highlight an interesting behavior of existing models, it fits into this package!
New tasks should be added to the shifthappens.tasks module.
Please refer to the API documentation for more details, as well as minimal examples. Moreover, inside the examples folder you can find implementations example implementations of tasks for the benchmark.
Adding models
Models should be added to the shifthappens.models module. An example implementation for wrapping a torchvision model is given in shifthappens.models.torchvision. Note that implementations are framework agnostic; further options include TensorFlow, and jax models, for which we will add example implementations soon.
License
All code in this repository is released under an Apache 2.0 license, which includes external contributions. All data used for creating new benchmarks should minimally be available without constrains for research purposes, and optionally free for commercial use as well.
Datasets without an explicit license statement will not be accepted into the benchmark.
1 maintainer
Metadata
1 maintainer
Release files for shifthappens 0.0.1a0
For a detailed explanation of source distributions (sdists) and built distributions (wheels), please see the package formats documentation.
Built distribution (wheel)
| File | Interpreter | ABI | Platform | Reset |
|---|---|---|---|---|
| shifthappens-0.0.1a0-py3-none-any.whl | Python 3 | none | any | Details |
Release files / shifthappens-0.0.1a0-py3-none-any.whl
| Download URL | shifthappens-0.0.1a0-py3-none-any.whl |
|---|---|
| Size | 29.4 kB |
| Tags | Python 3 |
|
SHA-256 checksum How to use checksums |
4cdab58e80c2064bca08b8fc3c4dd2ea7cb8abb8f7858b2ce72c95fee5463701
|
|
BLAKE2b-256 checksum How to use checksums |
d4deeb304b921dbe0a65091d8b689fe547419c3fd8acdaabbe74a701b4ee7a4f
|
| Upload date | |
|
Uploaded using Trusted Publishing? What is trusted publishing? |
No |
| Uploaded via |
twine/3.3.0 pkginfo/1.6.1 requests/2.24.0 setuptools/50.3.1.post20201107 requests-toolbelt/0.9.1 tqdm/4.50.2 CPython/3.8.5
|
