50k vs cifar10.train.npz

Author: w86763777

Readme.md

@@ -1,8 +1,8 @@
 # Collections of GANs
 
-Pytorch implementation of unsupervised GANs.
+Pytorch implementation of basic unsupervised GANs on CIFAR10.
 
-For more defails about calculating Inception Score and FID using pytorch can be found here [pytorch_gan_metrics](https://github.com/w86763777/pytorch-gan-metrics) 
+For more defails about calculating Inception Score and FID using pytorch can be found here [pytorch_gan_metrics](https://github.com/w86763777/pytorch-gan-metrics).
 
 ## Models
 - [x] DCGAN
@@ -18,15 +18,15 @@ For more defails about calculating Inception Score and FID using pytorch can be
     ```
 
 ## Results
-
+The FID is calculated by 50k generated images and CIFAR10 train set.
 |Model          |Dataset|Inception Score|FID  |
 |--------------|:-----:|:--------------:|:---:|
-|DCGAN          |CIFAR10|5.91(0.15)     |47.46|
-|WGAN(CNN)      |CIFAR10|6.46(0.24)     |38.98|
-|WGAN-GP(CNN)   |CIFAR10|7.69(0.19)     |22.81|
-|WGAN-GP(ResNet)|CIFAR10|7.80(0.20)     |21.48|
-|SNGAN(CNN)     |CIFAR10|7.64(0.21)     |21.86|
-|SNGAN(ResNet)  |CIFAR10|8.21(0.17)     |19.11|
+|DCGAN          |CIFAR10|6.01(0.05)     |42.72|
+|WGAN(CNN)      |CIFAR10|6.62(0.09)     |40.03|
+|WGAN-GP(CNN)   |CIFAR10|7.66(0.10)     |19.83|
+|WGAN-GP(ResNet)|CIFAR10|7.95(0.14)     |16.95|
+|SNGAN(CNN)     |CIFAR10|7.84(0.12)     |17.81|
+|SNGAN(ResNet)  |CIFAR10|8.31(0.10)     |14.32|
 
 ## Examples
 - DCGAN
@@ -54,13 +54,16 @@ For more defails about calculating Inception Score and FID using pytorch can be
     ![sngan_res_gif](https://drive.google.com/uc?export=view&id=1et3V7NbLEqH6aOWzkOQceNcnfY3WBOGz) ![sngan_res_png](https://drive.google.com/uc?export=view&id=1neYWCexP8kY2eixMpztNL50TKFLXZcBL)
 
 ## Reproduce
-- Download [cifar10.test.npz](https://drive.google.com/drive/folders/1UBdzl6GtNMwNQ5U-4ESlIer43tNjiGJC?usp=sharing) for calculating FID. Then, create folder `stats` for the npz files
+- Download [cifar10.train.npz](https://drive.google.com/drive/folders/1UBdzl6GtNMwNQ5U-4ESlIer43tNjiGJC?usp=sharing) for calculating FID. Then, create folder `stats` for the npz files
     ```
     stats
-    └── cifar10.test.npz
+    └── cifar10.train.npz
     ```
 
 - Train from scratch
+
+    Different methods are separated into different files for clear reading.
+
     ```bash
     # DCGAN
     python dcgan.py --flagfile ./configs/DCGAN_CIFAR10.txt
@@ -75,18 +78,18 @@ For more defails about calculating Inception Score and FID using pytorch can be
     # SNGAN(ResNet)
     python sngan.py --flagfile ./configs/SNGAN_CIFAR10_RES.txt
     ```
-    Though the training procedures of different GANs are almost identical, I still separate different methods into different files for clear reading.
+    
 
-## Learning curve
+## Learning Curves
 ![inception_score_curve](https://drive.google.com/uc?export=view&id=12JTJS5--2dDjFyVhHJ-b264Qp3S-v8xS)
 ![fid_curve](https://drive.google.com/uc?export=view&id=1P4e_DEyW4wvFubPSu5t_i2gVRoecGqs5)
 
 ## Change Log
 - 2021-01-10
     - Update pytorch to 1.10.1 and CUDA 11.3
-    - Calculate FID and Inception by `pytorch_gan_metrics`
+    - Use `pytorch_gan_metrics` to calculate FID and Inception Score
     - Use 50k generated images and CIFAR10 train set to calculate FID
-    - Fix default parameters
+    - Fix default parameters especially for `wgan.py`
 
 - 2021-04-16
     - Update pytorch to 1.8.1

configs/DCGAN_CIFAR10.txt

@@ -1,13 +1,13 @@
 --arch=cnn32
 --batch_size=128
 --dataset=cifar10
---fid_cache=./stats/cifar10.test.npz
+--fid_cache=./stats/cifar10.train.npz
 --logdir=./logs/DCGAN_CIFAR10
 --loss=bce
 --lr_D=0.0002
 --lr_G=0.0002
 --n_dis=1
---num_images=10000
+--num_images=50000
 --record
 --sample_step=500
 --sample_size=64

configs/SNGAN_CIFAR10_CNN.txt

@@ -1,13 +1,13 @@
 --arch=cnn32
 --batch_size=128
 --dataset=cifar10
---fid_cache=./stats/cifar10.test.npz
+--fid_cache=./stats/cifar10.train.npz
 --logdir=./logs/SNGAN_CIFAR10_CNN
 --loss=hinge
 --lr_D=0.0002
 --lr_G=0.0002
 --n_dis=1
---num_images=10000
+--num_images=50000
 --record
 --sample_step=500
 --sample_size=64

configs/SNGAN_CIFAR10_RES.txt

@@ -1,13 +1,13 @@
 --arch=res32
 --batch_size=64
 --dataset=cifar10
---fid_cache=./stats/cifar10.test.npz
+--fid_cache=./stats/cifar10.train.npz
 --logdir=./logs/SNGAN_CIFAR10_RES
 --loss=hinge
 --lr_D=0.0002
 --lr_G=0.0002
 --n_dis=5
---num_images=10000
+--num_images=50000
 --record
 --sample_step=500
 --sample_size=64

configs/WGANGP_CIFAR10_CNN.txt

@@ -2,13 +2,13 @@
 --arch=cnn32
 --batch_size=128
 --dataset=cifar10
---fid_cache=./stats/cifar10.test.npz
+--fid_cache=./stats/cifar10.train.npz
 --logdir=./logs/WGANGP_CIFAR10_CNN
 --loss=was
 --lr_D=0.0002
 --lr_G=0.0002
 --n_dis=1
---num_images=10000
+--num_images=50000
 --record
 --sample_step=500
 --sample_size=64

configs/WGANGP_CIFAR10_RES.txt

@@ -2,13 +2,13 @@
 --arch=res32
 --batch_size=64
 --dataset=cifar10
---fid_cache=./stats/cifar10.test.npz
+--fid_cache=./stats/cifar10.train.npz
 --logdir=./logs/WGANGP_CIFAR10_RES
 --loss=was
 --lr_D=0.0002
 --lr_G=0.0002
 --n_dis=5
---num_images=10000
+--num_images=50000
 --record
 --sample_step=500
 --sample_size=64

configs/WGAN_CIFAR10_CNN.txt

@@ -2,13 +2,13 @@
 --arch=cnn32
 --batch_size=128
 --dataset=cifar10
---fid_cache=./stats/cifar10.test.npz
+--fid_cache=./stats/cifar10.train.npz
 --logdir=./logs/WGAN_CIFAR10_CNN
 --loss=was
 --lr_D=0.0002
 --lr_G=0.0002
 --n_dis=1
---num_images=10000
+--num_images=50000
 --record
 --sample_step=500
 --sample_size=64

dcgan.py

@@ -48,12 +48,12 @@
 flags.DEFINE_integer('sample_size', 64, "sampling size of images")
 flags.DEFINE_string('logdir', './logs/DCGAN_CIFAR10', 'logging folder')
 flags.DEFINE_bool('record', True, "record inception score and FID")
-flags.DEFINE_string('fid_cache', './stats/cifar10.test.npz', 'FID cache')
+flags.DEFINE_string('fid_cache', './stats/cifar10.train.npz', 'FID cache')
 # generate
 flags.DEFINE_bool('generate', False, 'generate images')
 flags.DEFINE_string('pretrain', None, 'path to test model')
 flags.DEFINE_string('output', './outputs', 'path to output dir')
-flags.DEFINE_integer('num_images', 10000, 'the number of generated images')
+flags.DEFINE_integer('num_images', 50000, 'the number of generated images')
 
 device = torch.device('cuda:0')
 

wgan.py

@@ -51,12 +51,12 @@
 flags.DEFINE_integer('sample_size', 64, "sampling size of images")
 flags.DEFINE_string('logdir', './logs/WGAN_CIFAR10_CNN', 'logging folder')
 flags.DEFINE_bool('record', True, "record inception score and FID")
-flags.DEFINE_string('fid_cache', './stats/cifar10.test.npz', 'FID cache')
+flags.DEFINE_string('fid_cache', './stats/cifar10.train.npz', 'FID cache')
 # generate
 flags.DEFINE_bool('generate', False, 'generate images')
 flags.DEFINE_string('pretrain', None, 'path to test model')
 flags.DEFINE_string('output', './outputs', 'path to output dir')
-flags.DEFINE_integer('num_images', 10000, 'the number of generated images')
+flags.DEFINE_integer('num_images', 50000, 'the number of generated images')
 
 device = torch.device('cuda:0')
 

wgangp.py

@@ -51,12 +51,12 @@
 flags.DEFINE_integer('sample_size', 64, "sampling size of images")
 flags.DEFINE_string('logdir', './logs/WGANGP_CIFAR10_RES', 'logging folder')
 flags.DEFINE_bool('record', True, "record inception score and FID")
-flags.DEFINE_string('fid_cache', './stats/cifar10.test.npz', 'FID cache')
+flags.DEFINE_string('fid_cache', './stats/cifar10.train.npz', 'FID cache')
 # generate
 flags.DEFINE_bool('generate', False, 'generate images')
 flags.DEFINE_string('pretrain', None, 'path to test model')
 flags.DEFINE_string('output', './outputs', 'path to output dir')
-flags.DEFINE_integer('num_images', 10000, 'the number of generated images')
+flags.DEFINE_integer('num_images', 50000, 'the number of generated images')
 
 device = torch.device('cuda:0')