Ecosyste.ms: Awesome
An open API service indexing awesome lists of open source software.
https://github.com/bethgelab/model-vs-human
Benchmark your model on out-of-distribution datasets with carefully collected human comparison data (NeurIPS 2021 Oral)
https://github.com/bethgelab/model-vs-human
benchmark pytorch robustness tensorflow toolbox
Last synced: about 2 months ago
JSON representation
Benchmark your model on out-of-distribution datasets with carefully collected human comparison data (NeurIPS 2021 Oral)
- Host: GitHub
- URL: https://github.com/bethgelab/model-vs-human
- Owner: bethgelab
- Created: 2021-06-04T10:49:20.000Z (over 3 years ago)
- Default Branch: master
- Last Pushed: 2024-02-09T01:15:36.000Z (7 months ago)
- Last Synced: 2024-04-20T05:34:11.685Z (5 months ago)
- Topics: benchmark, pytorch, robustness, tensorflow, toolbox
- Language: Python
- Homepage: https://openreview.net/forum?id=QkljT4mrfs
- Size: 26.4 MB
- Stars: 319
- Watchers: 15
- Forks: 43
- Open Issues: 3
-
Metadata Files:
- Readme: README.md
- License: licenses/CODE_LICENSE
Awesome Lists containing this project
README
Benchmark •
Installation •
User experience •
Model zoo •
Datasets •
Credit & citation
# modelvshuman: Does your model generalise better than humans?
``modelvshuman`` is a Python toolbox to benchmark the gap between human and machine vision. Using this library, both PyTorch and TensorFlow models can be evaluated on 17 out-of-distribution datasets with high-quality human comparison data.
## :trophy: Benchmark
The top-10 models are listed here; training dataset size is indicated in brackets. Additionally, standard ResNet-50 is included as the last entry of the table for comparison. Model ranks are calculated across the full range of 52 models that we tested. If your model scores better than some (or even all) of the models here, please open a pull request and we'll be happy to include it here!
### Most human-like behaviour
winner | model | accuracy difference ↓ | observed consistency ↑ | error consistency ↑ | mean rank ↓ |
:----------------:| ------------------------------ |-----------------------------:|-----------------------------:|------------------------------:|-----------------------:|
:1st_place_medal: | [ViT-22B-384](https://arxiv.org/abs/2302.05442): ViT-22B (4B) | **.018** | **.783** | .258 | **1.67**|
:2nd_place_medal: | [CLIP](https://arxiv.org/abs/2103.00020): ViT-B (400M) | .023 | .758 | **.281** | 3|
:3rd_place_medal: | [ViT-22B-560](https://arxiv.org/abs/2302.05442): ViT-22B (4B) | .022 | .739 | **.281** | 3.33|
:clap: | [SWSL](https://arxiv.org/abs/1905.00546): ResNeXt-101 (940M) | .028 | .752 | .237 | 6|
:clap: | [BiT-M](https://arxiv.org/abs/1912.11370): ResNet-101x1 (14M) | .034 | .733 | .252 | 7|
:clap: | [BiT-M](https://arxiv.org/abs/1912.11370): ResNet-152x2 (14M) | .035 | .737 | .243 | 7.67|
:clap: | [ViT-L](https://openreview.net/forum?id=YicbFdNTTy) (1M) | .033 | .738 | .222 | 9.33|
:clap: | [BiT-M](https://arxiv.org/abs/1912.11370): ResNet-152x4 (14M) | .035 | .732 | .233 | 10.33|
:clap: | [BiT-M](https://arxiv.org/abs/1912.11370): ResNet-50x3 (14M) | .040 | .726 | .228 | 12|
:clap: | [ViT-L](https://openreview.net/forum?id=YicbFdNTTy) (14M) | .035 | .744 | .206 | 12|
... | standard [ResNet-50](https://arxiv.org/abs/1502.01852) (1M) | .087 | .665 | .208 | 31.33|
### Highest OOD (out-of-distribution) distortion robustness
winner | model | OOD accuracy ↑ | rank ↓ |
:----------------:| ----------------------------------------------------------------------------| -------------------------:|------------------:|
:1st_place_medal: | [ViT-22B-224](https://arxiv.org/abs/2302.05442): ViT-22B (4B) | **.837** | **1**|
:2nd_place_medal: | [Noisy Student](https://arxiv.org/abs/1911.04252): EfficientNet-L2 (300M) | .829 | 2|
:3rd_place_medal: | [ViT-22B-384](https://arxiv.org/abs/2302.05442): ViT-22B (4B) | .798 | 3|
:clap: | [ViT-L](https://openreview.net/forum?id=YicbFdNTTy) (14M) | .733 | 4|
:clap: | [CLIP](https://arxiv.org/abs/2103.00020): ViT-B (400M) | .708 | 5|
:clap: | [ViT-L](https://openreview.net/forum?id=YicbFdNTTy) (1M) | .706 | 6|
:clap: | [SWSL](https://arxiv.org/abs/1905.00546): ResNeXt-101 (940M) | .698 | 7|
:clap: | [BiT-M](https://arxiv.org/abs/1912.11370): ResNet-152x2 (14M) | .694 | 8|
:clap: | [BiT-M](https://arxiv.org/abs/1912.11370): ResNet-152x4 (14M) | .688 | 9|
:clap: | [BiT-M](https://arxiv.org/abs/1912.11370): ResNet-101x3 (14M) | .682 | 10|
... | standard [ResNet-50](https://arxiv.org/abs/1502.01852) (1M) | .559 | 34|
## :wrench: Installation
Simply clone the repository to a location of your choice and follow these steps (requires ``python3.8``):
1. Set the repository home path by running the following from the command line:
```
export MODELVSHUMANDIR=/absolute/path/to/this/repository/
```
2. Within the cloned repository, install package:
```
pip install -e .
```
(The -e option makes sure that changes to the code are reflected in the package, which is important e.g. if you add your own model or make any other changes)
## :microscope: User experience
Simply edit ``examples/evaluate.py`` as desired. This will test a list of models on out-of-distribution datasets, generating plots. If you then compile ``latex-report/report.tex``, all the plots will be included in one convenient PDF report.
## :camel: Model zoo
The following models are currently implemented:
- [x] 20+ standard supervised models from the [torchvision model zoo](https://pytorch.org/docs/1.4.0/torchvision/models.html)
- [x] 5 self-supervised contrastive models (InsDis, MoCo, MoCoV2, InfoMin, PIRL) from the [pycontrast repo](https://github.com/HobbitLong/PyContrast/)
- [x] 3 self-supervised contrastive SimCLR model variants (simclr_resnet50x1, simclr_resnet50x2, simclr_resnet50x4) from the [ptrnet repo](https://github.com/sacadena/ptrnets)
- [x] 3 vision transformer variants (vit_small_patch16_224, vit_base_patch16_224 and vit_large_patch16_224) from the [pytorch-image-models repo](https://github.com/rwightman/pytorch-image-models)
- [x] 10 adversarially "robust" models from [robust-models-transfer repo](https://github.com/microsoft/robust-models-transfer) implemented via the [ptrnet repo](https://github.com/sacadena/ptrnets)
- [x] 3 "ShapeNet" ResNet-50 models with different degree of stylized training from the [texture-vs-shape repo](https://github.com/rgeirhos/texture-vs-shape)
- [x] 3 BagNet models from the [BagNet repo](https://github.com/wielandbrendel/bag-of-feature-DNNs)
- [x] 1 semi-supervised ResNet-50 model pre-trained on 940M images from the [semi-supervised-ImageNet1K-models repo](https://github.com/facebookresearch/semi-supervised-ImageNet1K-models)
- [x] 6 Big Transfer models from the [pytorch-image-models repo](https://github.com/rwightman/pytorch-image-models)
If you e.g. add/implement your own model, please make sure to compute the ImageNet accuracy as a sanity check.
##### How to load a model
If you just want to load a model from the model zoo, this is what you can do:
```python
# loading a PyTorch model from the zoo
from modelvshuman.models.pytorch.model_zoo import InfoMin
model = InfoMin("InfoMin")
# loading a Tensorflow model from the zoo
from modelvshuman.models.tensorflow.model_zoo import efficientnet_b0
model = efficientnet_b0("efficientnet_b0")
```
Then, if you have a custom set of images that you want to evaluate the model on, load those (in the example below, called ``images``) and evaluate via:
```python
output_numpy = model.forward_batch(images)
# by default, type(output) is numpy.ndarray, which can be converted to a tensor via:
output_tensor = torch.tensor(output_numpy)
```
However, if you simply want to run a model through the generalisation datasets provided by the toolbox, we recommend to check the section on User experience.
##### How to list all available models
All implemented models are registered by the model registry, which can then be used to list all available models of a certain framework with the following method:
```python
from modelvshuman import models
print(models.list_models("pytorch"))
print(models.list_models("tensorflow"))
```
##### How to add a new model
Adding a new model is possible for standard PyTorch and TensorFlow models. Depending on the framework (pytorch / tensorflow), open ``modelvshuman/models//model_zoo.py``. Here, you can add your own model with a few lines of code - similar to how you would load it usually. If your model has a custom model definition, create a new subdirectory called ``modelvshuman/models//my_fancy_model/fancy_model.py`` which you can then import from ``model_zoo.py`` via ``from .my_fancy_model import fancy_model``.
## :file_folder: Datasets
In total, 17 datasets with human comparison data collected under highly controlled laboratory conditions in the [Wichmannlab](http://www.wichmannlab.org) are available.
Twelve datasets correspond to parametric or binary image distortions. Top row: colour/grayscale, contrast, high-pass, low-pass (blurring), phase noise, power equalisation. Bottom row: opponent colour, rotation, Eidolon I, II and III, uniform noise.
The remaining five datasets correspond to the following nonparametric image manipulations: sketch, stylized, edge, silhouette, texture-shape cue conflict.
##### How to load a dataset
Similarly, if you're interested in just loading a dataset, you can do this via:
```python
from modelvshuman.datasets import sketch
dataset = sketch(batch_size=16, num_workers=4)
```
Note that the datasets aren't available after installing the toolbox just yet. Instead, they are automatically downloaded the first time a model is evaluated on the dataset (see ``examples/evaluate.py``).
##### How to list all available datasets
```python
from modelvshuman import datasets
print(list(datasets.list_datasets().keys()))
```
## :credit_card: Credit
Psychophysical data were collected by us in the vision laboratory of the [Wichmannlab](http://www.wichmannlab.org).
That said, we used existing image dataset sources. 12 datasets were obtained from [Generalisation in humans and deep neural networks](http://papers.nips.cc/paper/7982-generalisation-in-humans-and-deep-neural-networks.pdf). 4 datasets were obtained from [ImageNet-trained CNNs are biased towards texture; increasing shape bias improves accuracy and robustness](https://openreview.net/forum?id=Bygh9j09KX). Additionally, we used 1 dataset from [Learning Robust Global Representations by Penalizing Local Predictive Power](https://arxiv.org/abs/1905.13549) (sketch images from ImageNet-Sketch).
We thank all model authors and repository maintainers for providing the models described above.
### Citation
@inproceedings{geirhos2021partial,
title={Partial success in closing the gap between human and machine vision},
author={Geirhos, Robert and Narayanappa, Kantharaju and Mitzkus, Benjamin and Thieringer, Tizian and Bethge, Matthias and Wichmann, Felix A and Brendel, Wieland},
booktitle={{Advances in Neural Information Processing Systems 34}},
year={2021},
}