Comments (11)
This is quite hard to tell. Did you manage to run the code on our dataset (e.g., facades)?
from bicyclegan.
Yes its working on your dataset(edges2shoes).I will create some more dataset and check it.
from bicyclegan.
Cool. Maybe you have a broken image or 1 channel image in your dataset. But it can be other reasons.
from bicyclegan.
I have the same problem when training on my own dataset with both input and output channel = 1, and HW 256x256. I use the following bash:
python ./train.py --gpu_ids 0 --display_id 1 --dataroot /mnt/ssd/IVS_Data_All_20181213 --dataset_mode alphaL --alphaL_data_json_file_full_path /mnt/ssd/IV S_Data_All_20181213/all_combine_20181229.json --num_threads 0 --name alphaL_bicycle_gan_10_01_2019_15 --model bicycle_gan --netG unet_256 --direction AtoB --c heckpoints_dir ../checkpoints/alphaL/ --preprocess crop --load_size 480 --crop_size 384 --nz 8 --input_nc 1 --output_nc 1 --niter 100 --niter_decay 100 --save _epoch_freq 10 --use_dropout
Additionally, I notice that just as @riktimmondal, error raise at bicycle_gan_model.py, line 129, in backward_D() pred_real = netD(real)
from bicyclegan.
I will have a look next week. Sorry that I am busy with a paper deadline now.
from bicyclegan.
after some dig, I found the problem may caused by un-full mini-batch. After I set dataset length to even, (default batch-size=2), It seems the training goes well.
from bicyclegan.
Yes, the code divides a batch into two. I can add an assertion.
from bicyclegan.
I added a function to check if the mini-batch is good for training.
from bicyclegan.
Thanks. I have another question.
Dut to fake_B=net_G(A,z) and tanh is the last non-linear-layer, fake_B should be in range(-1, 1). However, the real_B has mean 0.5 and std 0.5. This different value range between fake_B and real_B makes net_D easy to tell their different. Do I understand correctly? Thanks again.
from bicyclegan.
It normalizes an image to [-1, 1]. See this post for more details.
from bicyclegan.
So, transforms.Normalize() normalize a image of range [0, 1] to [-1, 1].
Thanks.
from bicyclegan.
Related Issues (20)
- Why does not large batch size like 128, 256 work well?
- Hi, please help me.
- Very Large Images
- Why my LPIPS distance is larger than what your paper say? HOT 3
- Question about conditional_D implementation HOT 3
- Not clear in the difference between the two latent spaces predicted HOT 5
- Test on single images HOT 1
- Compute graph wrong and one question HOT 4
- test_before_push is a great rapid test file, but seems outdated? HOT 3
- Metric reporting HOT 1
- Question about Encoder in cLR-GAN HOT 1
- Question about generating fake_B_random
- Question: Do you need two separate discriminators? HOT 2
- Incorrect discriminator update for opt.use_same_D HOT 1
- Regarding Training your Own Images HOT 4
- How to train on large images?
- Is there <pix2pix+noise> model code that can be directly run? HOT 1
- diversity question
- TypeError: __init__() got an unexpected keyword argument 'nl_layer'
- Regarding Latent Space Interpolation
Recommend Projects
-
React
A declarative, efficient, and flexible JavaScript library for building user interfaces.
-
Vue.js
🖖 Vue.js is a progressive, incrementally-adoptable JavaScript framework for building UI on the web.
-
Typescript
TypeScript is a superset of JavaScript that compiles to clean JavaScript output.
-
TensorFlow
An Open Source Machine Learning Framework for Everyone
-
Django
The Web framework for perfectionists with deadlines.
-
Laravel
A PHP framework for web artisans
-
D3
Bring data to life with SVG, Canvas and HTML. 📊📈🎉
-
Recommend Topics
-
javascript
JavaScript (JS) is a lightweight interpreted programming language with first-class functions.
-
web
Some thing interesting about web. New door for the world.
-
server
A server is a program made to process requests and deliver data to clients.
-
Machine learning
Machine learning is a way of modeling and interpreting data that allows a piece of software to respond intelligently.
-
Visualization
Some thing interesting about visualization, use data art
-
Game
Some thing interesting about game, make everyone happy.
Recommend Org
-
Facebook
We are working to build community through open source technology. NB: members must have two-factor auth.
-
Microsoft
Open source projects and samples from Microsoft.
-
Google
Google ❤️ Open Source for everyone.
-
Alibaba
Alibaba Open Source for everyone
-
D3
Data-Driven Documents codes.
-
Tencent
China tencent open source team.
from bicyclegan.