Skip to content

PyTorch - FID calculation with proper image resizing and quantization steps [CVPR 2022]

License

Notifications You must be signed in to change notification settings

GaParmar/clean-fid

Folders and files

NameName
Last commit message
Last commit date

Latest commit

 
 
 
 
 
 
 
 
 
 
 
 
 
 
 

Repository files navigation

clean-fid for Evaluating Generative Models


Downloads Downloads

Project | Paper | Slides | Colab-FID | Colab-Resize | Leaderboard Tables
Quick start: Calculate FID | Calculate KID

[New] Computing the FID using CLIP features [Kynkäänniemi et al, 2022] is now supported. See here for more details.

The FID calculation involves many steps that can produce inconsistencies in the final metric. As shown below, different implementations use different low-level image quantization and resizing functions, the latter of which are often implemented incorrectly.

We provide an easy-to-use library to address the above issues and make the FID scores comparable across different methods, papers, and groups.

FID Steps


Corresponding Manuscript

On Aliased Resizing and Surprising Subtleties in GAN Evaluation
Gaurav Parmar, Richard Zhang, Jun-Yan Zhu
CVPR, 2022
CMU and Adobe

If you find this repository useful for your research, please cite the following work.

@inproceedings{parmar2021cleanfid,
  title={On Aliased Resizing and Surprising Subtleties in GAN Evaluation},
  author={Parmar, Gaurav and Zhang, Richard and Zhu, Jun-Yan},
  booktitle={CVPR},
  year={2022}
}


Aliased Resizing Operations

The definitions of resizing functions are mathematical and should never be a function of the library being used. Unfortunately, implementations differ across commonly-used libraries. They are often implemented incorrectly by popular libraries. Try out the different resizing implementations in the Google colab notebook here.


The inconsistencies among implementations can have a drastic effect of the evaluations metrics. The table below shows that FFHQ dataset images resized with bicubic implementation from other libraries (OpenCV, PyTorch, TensorFlow, OpenCV) have a large FID score (≥ 6) when compared to the same images resized with the correctly implemented PIL-bicubic filter. Other correctly implemented filters from PIL (Lanczos, bilinear, box) all result in relatively smaller FID score (≤ 0.75). Note that since TF 2.0, the new flag antialias (default: False) can produce results close to PIL. However, it was not used in the existing TF-FID repo and set as False by default.

JPEG Image Compression

Image compression can have a surprisingly large effect on FID. Images are perceptually indistinguishable from each other but have a large FID score. The FID scores under the images are calculated between all FFHQ images saved using the corresponding JPEG format and the PNG format.

Below, we study the effect of JPEG compression for StyleGAN2 models trained on the FFHQ dataset (left) and LSUN outdoor Church dataset (right). Note that LSUN dataset images were collected with JPEG compression (quality 75), whereas FFHQ images were collected as PNG. Interestingly, for LSUN dataset, the best FID score (3.48) is obtained when the generated images are compressed with JPEG quality 87.


Quick Start

  • install the library
    pip install clean-fid
    

Computing FID

  • Compute FID between two image folders
    from cleanfid import fid
    score = fid.compute_fid(fdir1, fdir2)
    
  • Compute FID between one folder of images and pre-computed datasets statistics (e.g., FFHQ)
    from cleanfid import fid
    score = fid.compute_fid(fdir1, dataset_name="FFHQ", dataset_res=1024, dataset_split="trainval70k")
    
  • Compute FID using a generative model and pre-computed dataset statistics:
    from cleanfid import fid
    # function that accepts a latent and returns an image in range[0,255]
    gen = lambda z: GAN(latent=z, ... , <other_flags>)
    score = fid.compute_fid(gen=gen, dataset_name="FFHQ",
            dataset_res=256, num_gen=50_000, dataset_split="trainval70k")
    

Computing CLIP-FID

To use the CLIP features when computing the FID [Kynkäänniemi et al, 2022], specify the flag model_name="clip_vit_b_32"

  • e.g. to compute the CLIP-FID between two folders of images use the following commands.
    from cleanfid import fid
    score = fid.compute_fid(fdir1, fdir2, mode="clean", model_name="clip_vit_b_32")
    

Computing KID

The KID score can be computed using a similar interface as FID. The dataset statistics for KID are only precomputed for smaller datasets AFHQ, BreCaHAD, and MetFaces.

  • Compute KID between two image folders
    from cleanfid import fid
    score = fid.compute_kid(fdir1, fdir2)
    
  • Compute KID between one folder of images and pre-computed datasets statistics
    from cleanfid import fid
    score = fid.compute_kid(fdir1, dataset_name="brecahad", dataset_res=512, dataset_split="train")
    
  • Compute KID using a generative model and pre-computed dataset statistics:
    from cleanfid import fid
    # function that accepts a latent and returns an image in range[0,255]
    gen = lambda z: GAN(latent=z, ... , <other_flags>)
    score = fid.compute_kid(gen=gen, dataset_name="brecahad", dataset_res=512, num_gen=50_000, dataset_split="train")
    

Supported Precomputed Datasets

We provide precompute statistics for the following commonly used configurations. Please contact us if you want to add statistics for your new datasets.

Task Dataset Resolution Reference Split # Reference Images mode
Image Generation cifar10 32 train 50,000 clean, legacy_tensorflow, legacy_pytorch
Image Generation cifar10 32 test 10,000 clean, legacy_tensorflow, legacy_pytorch
Image Generation ffhq 1024, 256 trainval 50,000 clean, legacy_tensorflow, legacy_pytorch
Image Generation ffhq 1024, 256 trainval70k 70,000 clean, legacy_tensorflow, legacy_pytorch
Image Generation lsun_church 256 train 50,000 clean, legacy_tensorflow, legacy_pytorch
Image Generation lsun_church 256 trainfull 126,227 clean
Image Generation lsun_horse 256 train 50,000 clean, legacy_tensorflow, legacy_pytorch
Image Generation lsun_horse 256 trainfull 2,000,340 clean
Image Generation lsun_cat 256 train 50,000 clean, legacy_tensorflow, legacy_pytorch
Image Generation lsun_cat 256 trainfull 1,657,264 clean, legacy_tensorflow, legacy_pytorch
Few Shot Generation afhq_cat 512 train 5153 clean, legacy_tensorflow, legacy_pytorch
Few Shot Generation afhq_dog 512 train 4739 clean, legacy_tensorflow, legacy_pytorch
Few Shot Generation afhq_wild 512 train 4738 clean, legacy_tensorflow, legacy_pytorch
Few Shot Generation brecahad 512 train 1944 clean, legacy_tensorflow, legacy_pytorch
Few Shot Generation metfaces 1024 train 1336 clean, legacy_tensorflow, legacy_pytorch
Image to Image horse2zebra 256 test 140 clean, legacy_tensorflow, legacy_pytorch
Image to Image cat2dog 256 test 500 clean, legacy_tensorflow, legacy_pytorch

Using precomputed statistics In order to compute the FID score with the precomputed dataset statistics, use the corresponding options. For instance, to compute the clean-fid score on generated 256x256 FFHQ images use the command:

fid_score = fid.compute_fid(fdir1, dataset_name="ffhq", dataset_res=256,  mode="clean", dataset_split="trainval70k")

Create Custom Dataset Statistics

  • dataset_path: folder where the dataset images are stored

  • custom_name: name to be used for the statistics

  • Generating custom statistics (saved to local cache)

    from cleanfid import fid
    fid.make_custom_stats(custom_name, dataset_path, mode="clean")
    
  • Using the generated custom statistics

    from cleanfid import fid
    score = fid.compute_fid("folder_fake", dataset_name=custom_name,
              mode="clean", dataset_split="custom")
    
  • Removing the custom stats

    from cleanfid import fid
    fid.remove_custom_stats(custom_name, mode="clean")
    
  • Check if a custom statistic already exists

    from cleanfid import fid
    fid.test_stats_exists(custom_name, mode)
    

Backwards Compatibility

We provide two flags to reproduce the legacy FID score.

  • mode="legacy_pytorch"
    This flag is equivalent to using the popular PyTorch FID implementation provided here
    The difference between using clean-fid with this option and code is ~2e-06
    See doc for how the methods are compared

  • mode="legacy_tensorflow"
    This flag is equivalent to using the official implementation of FID released by the authors.
    The difference between using clean-fid with this option and code is ~2e-05
    See doc for detailed steps for how the methods are compared


Building clean-fid locally from source

python setup.py bdist_wheel
pip install dist/*

CleanFID Leaderboard for common tasks

We compute the FID scores using the corresponding methods used in the original papers and using the Clean-FID proposed here. All values are computed using 10 evaluation runs. We provide an API to query the results shown in the tables below directly from the pip package.

If you would like to add new numbers and models to our leaderboard, feel free to contact us.

CIFAR-10 (few shot)

The test set is used as the reference distribution and compared to 10k generated images.

100% data (unconditional)

Model Legacy-FID
(reported)
Legacy-FID
(reproduced)
Clean-FID
stylegan2 (+ada + tuning) [Karras et al, 2020] - † - † 8.20 ± 0.10
stylegan2 (+ada) [Karras et al, 2020] - † - † 9.26 ± 0.06
stylegan2 (diff-augment) [Zhao et al, 2020] [ckpt] 9.89 9.90 ± 0.09 10.85 ± 0.10
stylegan2 (mirror-flips) [Karras et al, 2020] [ckpt] 11.07 11.07 ± 0.10 12.96 ± 0.07
stylegan2 (without-flips) [Karras et al, 2020] - † - † 14.53 ± 0.13
AutoGAN (config A) [Gong et al, 2019] - † - † 21.18 ± 0.12
AutoGAN (config B) [Gong et al, 2019] - † - † 22.46 ± 0.15
AutoGAN (config C) [Gong et al, 2019] - † - † 23.62 ± 0.30

† These methods use the training set as the reference distribution and compare to 50k generated images

20% data

Model Legacy-FID
(reported)
Legacy-FID
(reproduced)
Clean-FID
stylegan2-diff-augment [Zhao et al, 2020] [ckpt] 12.15 12.12 ± 0.15 14.18 ± 0.13
stylegan2-mirror-flips [Karras et al, 2020] [ckpt] 23.08 23.01 ± 0.19 29.49 ± 0.17

10% data

Model Legacy-FID
(reported)
Legacy-FID
(reproduced)
Clean-FID
stylegan2-diff-augment [Zhao et al, 2020] [ckpt] 14.50 14.53 ± 0.12 16.98 ± 0.18
stylegan2-mirror-flips [Karras et al, 2020] [ckpt] 36.02 35.94 ± 0.17 43.60 ± 0.17

CIFAR-100 (few shot)

The test set is used as the reference distribution and compared to 10k generated images.

100% data

Model Legacy-FID
(reported)
Legacy-FID
(reproduced)
Clean-FID
stylegan2-mirror-flips [Karras et al, 2020] [ckpt] 16.54 16.44 ± 0.19 18.44 ± 0.24
stylegan2-diff-augment [Zhao et al, 2020] [ckpt] 15.22 15.15 ± 0.13 16.80 ± 0.13

20% data

Model Legacy-FID
(reported)
Legacy-FID
(reproduced)
Clean-FID
stylegan2-mirror-flips [Karras et al, 2020] [ckpt] 32.30 32.26 ± 0.19 34.88 ± 0.14
stylegan2-diff-augment [Zhao et al, 2020] [ckpt] 16.65 16.74 ± 0.10 18.49 ± 0.08

10% data

Model Legacy-FID
(reported)
Legacy-FID
(reproduced)
Clean-FID
stylegan2-mirror-flips [Karras et al, 2020] [ckpt] 45.87 45.97 ± 0.20 46.77 ± 0.19
stylegan2-diff-augment [Zhao et al, 2020] [ckpt] 20.75 20.69 ± 0.12 23.40 ± 0.09

FFHQ

all images @ 1024x1024
Values are computed using 50k generated images

Model Legacy-FID
(reported)
Legacy-FID
(reproduced)
Clean-FID Reference Split
stylegan1 (config A) [Karras et al, 2020] 4.4 4.39 ± 0.03 4.77 ± 0.03 trainval
stylegan2 (config B) [Karras et al, 2020] 4.39 4.43 ± 0.03 4.89 ± 0.03 trainval
stylegan2 (config C) [Karras et al, 2020] 4.38 4.40 ± 0.02 4.79 ± 0.02 trainval
stylegan2 (config D) [Karras et al, 2020] 4.34 4.34 ± 0.02 4.78 ± 0.03 trainval
stylegan2 (config E) [Karras et al, 2020] 3.31 3.33 ± 0.02 3.79 ± 0.02 trainval
stylegan2 (config F) [Karras et al, 2020] [ckpt] 2.84 2.83 +- 0.03 3.06 +- 0.02 trainval
stylegan2 [Karras et al, 2020] [ckpt] N/A 2.76 ± 0.03 2.98 ± 0.03 trainval70k

140k - images @ 256x256 (entire training set with horizontal flips) The 70k images from trainval70k set is used as the reference images and compared to 50k generated images.

Model Legacy-FID
(reported)
Legacy-FID
(reproduced)
Clean-FID
zCR [Zhao et al, 2020] 3.45 ± 0.19 3.29 ± 0.01 3.40 ± 0.01
stylegan2 [Karras et al, 2020] 3.66 ± 0.10 3.57 ± 0.03 3.73 ± 0.03
PA-GAN [Zhang and Khoreva et al, 2019] 3.78 ± 0.06 3.67 ± 0.03 3.81 ± 0.03
stylegan2-ada [Karras et al, 2020] 3.88 ± 0.13 3.84 ± 0.02 3.93 ± 0.02
Auxiliary rotation [Chen et al, 2019] 4.16 ± 0.05 4.10 ± 0.02 4.29 ± 0.03
Adaptive Dropout [Karras et al, 2020] 4.16 ± 0.05 4.09 ± 0.02 4.20 ± 0.02
Spectral Norm [Miyato et al, 2018] 4.60 ± 0.19 4.43 ± 0.02 4.65 ± 0.02
WGAN-GP [Gulrajani et al, 2017] 6.54 ± 0.37 6.19 ± 0.03 6.62 ± 0.03

† reported by [Karras et al, 2020]

30k - images @ 256x256 (Few Shot Generation)
The 70k images from trainval70k set is used as the reference images and compared to 50k generated images.

Model Legacy-FID
(reported)
Legacy-FID
(reproduced)
Clean-FID
stylegan2 [Karras et al, 2020] [ckpt] 6.16 6.14 ± 0.064 6.49 ± 0.068
DiffAugment-stylegan2 [Zhao et al, 2020] [ckpt] 5.05 5.07 ± 0.030 5.18 ± 0.032

10k - images @ 256x256 (Few Shot Generation)
The 70k images from trainval70k set is used as the reference images and compared to 50k generated images.

Model Legacy-FID
(reported)
Legacy-FID
(reproduced)
Clean-FID
stylegan2 [Karras et al, 2020] [ckpt] 14.75 14.88 ± 0.070 16.04 ± 0.078
DiffAugment-stylegan2 [Zhao et al, 2020] [ckpt] 7.86 7.82 ± 0.045 8.12 ± 0.044

5k - images @ 256x256 (Few Shot Generation)
The 70k images from trainval70k set is used as the reference images and compared to 50k generated images.

Model Legacy-FID
(reported)
Legacy-FID
(reproduced)
Clean-FID
stylegan2 [Karras et al, 2020] [ckpt] 26.60 26.64 ± 0.086 28.17 ± 0.090
DiffAugment-stylegan2 [Zhao et al, 2020] [ckpt] 10.45 10.45 ± 0.047 10.99 ± 0.050

1k - images @ 256x256 (Few Shot Generation)
The 70k images from trainval70k set is used as the reference images and compared to 50k generated images.

Model Legacy-FID
(reported)
Legacy-FID
(reproduced)
Clean-FID
stylegan2 [Karras et al, 2020] [ckpt] 62.16 62.14 ± 0.108 64.17 ± 0.113
DiffAugment-stylegan2 [Zhao et al, 2020] [ckpt] 25.66 25.60 ± 0.071 27.26 ± 0.077

LSUN Categories

100% data
The 50k images from train set is used as the reference images and compared to 50k generated images.

Category Model Legacy-FID
(reported)
Legacy-FID
(reproduced)
Clean-FID
Outdoor Churches stylegan2 [Karras et al, 2020] [ckpt] 3.86 3.87 ± 0.029 4.08 ± 0.028
Horses stylegan2 [Karras et al, 2020] [ckpt] 3.43 3.41 ± 0.021 3.62 ± 0.023
Cat stylegan2 [Karras et al, 2020] [ckpt] 6.93 7.02 ± 0.039 7.47 ± 0.035

LSUN CAT - 30k images (Few Shot Generation)
All 1,657,264 images from trainfull split are used as the reference images and compared to 50k generated images.

Model Legacy-FID
(reported)
Legacy-FID
(reproduced)
Clean-FID
stylegan2-mirror-flips [Karras et al, 2020] [ckpt] 10.12 10.15 ± 0.04 10.87 ± 0.04
stylegan2-diff-augment [Zhao et al, 2020] [ckpt] 9.68 9.70 ± 0.07 10.25 ± 0.07

LSUN CAT - 10k images (Few Shot Generation)
All 1,657,264 images from trainfull split are used as the reference images and compared to 50k generated images.

Model Legacy-FID
(reported)
Legacy-FID
(reproduced)
Clean-FID
stylegan2-mirror-flips [Karras et al, 2020] [ckpt] 17.93 17.98 ± 0.09 18.71 ± 0.09
stylegan2-diff-augment [Zhao et al, 2020] [ckpt] 12.07 12.04 ± 0.08 12.53 ± 0.08

LSUN CAT - 5k images (Few Shot Generation)
All 1,657,264 images from trainfull split are used as the reference images and compared to 50k generated images.

Model Legacy-FID
(reported)
Legacy-FID
(reproduced)
Clean-FID
stylegan2-mirror-flips [Karras et al, 2020] [ckpt] 34.69 34.66 ± 0.12 35.85 ± 0.12
stylegan2-diff-augment [Zhao et al, 2020] [ckpt] 16.11 16.11 ± 0.09 16.79 ± 0.09

LSUN CAT - 1k images (Few Shot Generation)
All 1,657,264 images from trainfull split are used as the reference images and compared to 50k generated images.

Model Legacy-FID
(reported)
Legacy-FID
(reproduced)
Clean-FID
stylegan2-mirror-flips [Karras et al, 2020] [ckpt] 182.85 182.80 ± 0.21 185.86 ± 0.21
stylegan2-diff-augment [Zhao et al, 2020] [ckpt] 42.26 42.07 ± 0.16 43.12 ± 0.16

AFHQ (Few Shot Generation)

AFHQ Dog
All 4739 images from train split are used as the reference images and compared to 50k generated images.

Model Legacy-FID
(reported)
Legacy-FID
(reproduced)
Clean-FID
stylegan2 [Karras et al, 2020] [ckpt] 19.37 19.34 ± 0.08 20.10 ± 0.08
stylegan2-ada [Karras et al, 2020] [ckpt] 7.40 7.41 ± 0.02 7.61 ± 0.02

AFHQ Wild
All 4738 images from train split are used as the reference images and compared to 50k generated images.

Model Legacy-FID
(reported)
Legacy-FID
(reproduced)
Clean-FID
stylegan2 [Karras et al, 2020] [ckpt] 3.48 3.55 ± 0.03 3.66 ± 0.02
stylegan2-ada [Karras et al, 2020] [ckpt] 3.05 3.01 ± 0.02 3.03 ± 0.02

BreCaHAD (Few Shot Generation)

All 1944 images from train split are used as the reference images and compared to 50k generated images.

Model Legacy
FID
(reported)
Legacy
FID
(reproduced)
Clean-FID Legacy
KID
(reported)
10^3
Legacy
KID
(reproduced)
10^3
Clean
KID
10^3
stylegan2 [Karras et al, 2020] [ckpt] 97.72 97.46 ± 0.17 98.35 ± 0.17 89.76 89.90 ± 0.31 92.51 ± 0.32
stylegan2-ada [Karras et al, 2020] [ckpt] 15.71 15.70 ± 0.06 15.63 ± 0.06 2.88 2.93 ± 0.08 3.08 ± 0.08

MetFaces (Few Shot Generation)

All 1336 images from train split are used as the reference images and compared to 50k generated images.

Model Legacy
FID
(reported)
Legacy
FID
(reproduced)
Clean-FID Legacy
KID
(reported)
10^3
Legacy
KID
(reproduced)
10^3
Clean
KID
10^3
stylegan2 [Karras et al, 2020] [ckpt] 57.26 57.36 ± 0.10 65.74 ± 0.11 35.66 35.69 ± 0.16 40.90 ± 0.14
stylegan2-ada [Karras et al, 2020] [ckpt] 18.22 18.18 ± 0.03 19.60 ± 0.03 2.41 2.38 ± 0.05 2.86 ± 0.04

Horse2Zebra (Image to Image Translation)

All 140 images from test split are used as the reference images and compared to 120 translated images.

Model Legacy-FID
(reported)
Legacy-FID
(reproduced)
Clean-FID
CUT [Park et al, 2020] 45.5 45.51 43.71
Distance [Benaim and Wolf et al, 2017] reported by [Park et al, 2020] 72.0 71.97 71.01
FastCUT [Park et al, 2020] 73.4 73.38 72.53
CycleGAN [Zhu et al, 2017] reported by [Park et al, 2020] 77.2 77.20 75.17
SelfDistance [Benaim and Wolf et al, 2017] reported by [Park et al, 2020] 80.8 80.78 79.28
GCGAN [Fu et al, 2019] reported by [Park et al, 2020] 86.7 85.86 83.65
MUNIT [Huang et al, 2018] reported by [Park et al, 2020] 133.8 - † 120.48
DRIT [Lee et al, 2017] reported by [Park et al, 2020] 140.0 - † 99.56

† The translated images for these methods were intitially compared by [Park et al, 2020] using .jpeg compression. We retrain these two methods using the same protocal and generate the images as .png for a fair comparision.


Cat2Dog (Image to Image Translation)

All 500 images from test split are used as the reference images and compared to 500 translated images.

Model Legacy-FID
(reported)
Legacy-FID
(reproduced)
Clean-FID
CUT [Park et al, 2020] 76.2 76.21 77.58
FastCUT [Park et al, 2020] 94.0 93.95 95.37
GCGAN [Fu et al, 2019] reported by [Park et al, 2020] 96.6 96.61 96.49
MUNIT [Huang et al, 2018] reported by [Park et al, 2020] 104.4 - † 123.73
DRIT [Lee et al, 2017] reported by [Park et al, 2020] 123.4 - † 127.21
SelfDistance [Benaim and Wolf et al, 2017] reported by [Park et al, 2020] 144.4 144.42 147.23
Distance [Benaim and Wolf et al, 2017] reported by [Park et al, 2020] 155.3 155.34 158.39

† The translated images for these methods were intitially compared by [Park et al, 2020] using .jpeg compression. We retrain these two methods using the same protocal and generate the images as .png for a fair comparision.


Related Projects

torch-fidelity: High-fidelity performance metrics for generative models in PyTorch.
TTUR: Two time-scale update rule for training GANs.
LPIPS: Perceptual Similarity Metric and Dataset.


Licenses

All material in this repository is made available under the MIT License.

inception_pytorch.py is derived from the PyTorch implementation of FID provided by Maximilian Seitzer. These files were originally shared under the Apache 2.0 License.

inception-2015-12-05.pt is a torchscript model of the pre-trained Inception-v3 network by Christian Szegedy, Vincent Vanhoucke, Sergey Ioffe, Jonathon Shlens, and Zbigniew Wojna. The network was originally shared under Apache 2.0 license on the TensorFlow Models repository. The torchscript wrapper is provided by Tero Karras and Miika Aittala and Janne Hellsten and Samuli Laine and Jaakko Lehtinen and Timo Aila which is released under the Nvidia Source Code License.