May I ask what kind of value range will your losses eventually stabilize in, including d1, d2, ect?
The value g_local loss has remained steady at about 3 when I try to train the Drive Dataset from 18 to 200 epochs.
I wonder what's the reason for that

Dear author：
URLs of pre-trained weights are invalid

Hi,
When I am training the model with the newer version of tf2.6, I got this warning message:

But my training seems goes well, don't know if this is a bug.

When I run train.py, I found the loss is 'nan' at behind of 2-epochs. Do you have this problem when you train. So I want to know why is it and how to solve this problem.

Hi, great stuff here, really interesting approach!

Are there pre-trained models weights available somewhere? I have an application in mind that I would love to use them for, would be citing your paper.

Hi, I got this error running train.py using DRIVE dataset in Colab:

Traceback (most recent call last):
File "/content/drive/MyDrive/RVGAN-master/train.py", line 215, in
train(d_model1, d_model2,g_model_coarse, g_model_fine, rvgan_model, dataset, n_epochs=args.epochs, n_batch=args.batch_size, n_patch=[128,64],savedir=args.savedir)
File "/content/drive/MyDrive/RVGAN-master/train.py", line 100, in train
g_global_loss,_ = g_global_model.train_on_batch([X_realA_half,X_realB_half], [X_realC_half])

ValueError: The two structures don't have the same sequence length. Input structure has length 1, while shallow structure has length 2.

Would you help me to fix it?

Hi, I tried training the model on tf 2.6.0 and some problems happens. How should i fix this?

Hello, the pretrained model of CHASE can't be loaded. Error message is
ValueError: Shapes (7, 7, 4, 128) and (64, 4, 7, 7) are incompatible.

In addition, could you provide the trained model?
I have tried many times, but I can't get the f1 score near which descaribed in the paper.
Until now, the best F1 score trained on drive is 0.78, Se is 0.74. The results on STARE is better(f1=0.8030 Se=0.8191). The training process is time-consuming(I use a card of 2080ti).

Hi,

I got this error. I tried to drop batch size, it didn't work.

System features as below:

Thank you

Hello! can you send me your generated dataset npz file? I can't generate is using your code

Hi
when i run infer.py, an error occurred. I had installed libtiff,
Traceback (most recent call last):
File "infer.py", line 7, in
from libtiff import TIFF
File "/home/miniconda3/envs/gwc_test/lib/python3.6/site-packages/libtiff/init.py", line 23, in
from .libtiff_ctypes import libtiff, TIFF, TIFF3D # noqa: F401
File "/home/miniconda3/envs/gwc_test/lib/python3.6/site-packages/libtiff/libtiff_ctypes.py", line 128, in
value = eval(value)
File "", line 1
\

looking forward to your reply
best wishes

Do we have the code for printing out the metrics like auc, F1 score and so on?

I am trying to train the DRIVE data with RVGAN-tf-2.6. An issue I'm finding is the predictions in local_plot are always blank even after 53 epochs of training which took over 20 hours:

local_plot_000053.png

The global_plot though is visible

global_plot_000053.png

I am trying to reproduce eval.py IOU values but the predictions are always blank and looks like it is because predictions from g_local_model are always blank though not from g_global_model

Hi,
I noticed that in your paper the inference time for each image is 0.025 second, is that for patch image or the whole image?

Hi,
　I downloaded your code and run it, but I can't get your results. Do I need to pay attention to certain details during the training process? And is your model easy to train?
Looking forward to your reply.

Hi!Thank you for your work！
I'm a little confused that you said，we should looped over all 100 saved weights to find the best performing coarse and fine generator pairs and train again.Which data do we use to determine the best model？

I have read your paper and run your code on my computer. I found some issues.

1. SFA block written in the code is different from which described in the paper. It's short of a input added to the terminal output.
2. Discriminator Residual Block described in the paper is not used in your code. In the code, Discriminator Residual Block just outputs two times of input.
   And i have a question.
   Is the network in the code no need to modify identical with which in the paper?
   Look forward to your reply!

I meet i problem. I don't understand you how to solve the data number is small,and how to get the number 4320 for STARE、15120 for CHASE-DB1、4200 for DRIVE .can you explain it? or send me your preprocess datasets. Thanks!!!

After running train.py i am having nan value for all the losses.
Code is running on:
tensorflow-gpu = 2.0.3
GPU - 3090 Ti

Could you provide the data set of npz format or provide preprocessed code？

I try to do your work,but meet a bug.



My tensorflow is 2.7.4 and keras is 2.7.0. I don't know what whould I do.

The paper descire the model and training skill very well in detail, but it seems to forget to describe the generation of the final output image. How do you use the results from Gf and Gc to predict an image?