Comments (11)
OK, I will try, thank you very much! And I just find the result of the 200th epoch on the training set is pretty good and the performance of 600th epoch is poor.
from instagan.
Great! Maybe 600 epoch is too much for low-resolution images (overfitting?). Early stopping could be a solution :)
from instagan.
Hi, did you trained InstaGAN model with predicted masks?
In my experience, the performance of InstaGAN was pretty sensitive to the mask qualities. Hence, although the results of pix2pix model is "pretty good", it may not enough for training of InstaGAN.
In my experiments, I trained InstaGAN with "real masks" and only used predicted masks for inference. Improving the robustness to the mask quality would be an interesting research direction.
from instagan.
Thank you. I train the instagan with real mask, but test it with the predict masks.
from instagan.
I used real masks to train the instagan model by running: python train.py --dataroot ./datasets/jeans2skirts_ccp_crop --model insta_gan --name jeans2skirts_ccp_crop_instagan --loadSizeH 200 --loadSizeW 80 --fineSizeH 200 --fineSizeW 80 --niter 400 --niter_decay 200 --batch_size 3 --gpu_ids 0,1,2. And I just test it with real mask of jeans and skirts, and the result is very bad either.
from instagan.
Hi, how about performance on your training set? Can you try with the pretrained network?
Also, as the training set and test set are different, although it performs okay for training set, it may fail for test set. In my experience, front-looking images were easier to transform.
from instagan.
Hi, performance on on my training set is bad either. The perfomance of pretrain network is good on the CCP dataset, but not good enough on some low pixel photos.
from instagan.
As the pretrained networks are trained on high-resolution images, it'd be better to match the resolution. How about scale up your low-resolution images and use pretrained networks?
from instagan.
That's a good idea. But I want to apply the model to the low resolution photos, so I compressed the image of the CCP data set and tried to train to instaGAN model with the real mask and those low resolution photos. However the trainning result were not satisfied.
from instagan.
Why not translate high-resolution images and resize it to low-resolution?
Maybe the current model parameter is too big for low-resolution images. You can try smaller models.
from instagan.
Yeah, that's what I think.
from instagan.
Related Issues (20)
- hi, I did an experiments on hair style transfer, but it seem don't work. HOT 11
- dataloader is broken HOT 1
- IndexError: list index out of range HOT 1
- broadcasting error (output) HOT 1
- Two inputs' ranges in models/insta_gan_model.py 230 row
- where can I download the pre-train model? HOT 2
- AttributeError: InstaGANModel instance has no attribute 'fake_B_img_sng'
- From my training logs, the TEACHER is always 0. Is that normal?
- trouble in datsets create
- Different Amount of Maks in training and test time HOT 1
- RuntimeError: output with shape [1, 240, 160] doesn't match the broadcast shape [3, 240, 160] HOT 1
- The design of adding mask to CycleGAN HOT 1
- RuntimeError: output with shape [1, 300, 200] doesn't match the broadcast shape [3, 300, 200] HOT 2
- Can InstaGAN be used for minor attribute manipulation? HOT 1
- While testing also the bnary masks are needed for bothe the domains?
- What happened to the (pear2orange) generated image without deformation
- Convert only one
- my own groundtruth images
- Calculation of masked classification score
- RuntimeError: cublas runtime error during training
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 instagan.