OK, I have finally run your scripts well, but I'm still confused about the training annotations, why is that trainannots are all black and grey images, I have tried to print out the values, but only 1 and nothing.
How do I train my own datasets, with rgb image how to convert its mask, thank you.

Data's link is 404 error

Hello ShawDa,

Thanks for sharing your code! I've been playing around with UNet in Python. I've been wondering how you manage to train and test on images that aren't perfect squares? If I send in a not-square input image, I'll get a squashed square output. Of course, resizing it doesn't help much...

Any insight you have to share is appreciated. Thanks!

This is not really an issue, but more of a request :) You write you made an implementation of unet-rgb for satellite imagery. Is this a repository you'd consider sharing?

how to place dataset? I have database containing 20 rgb images and respective masks. Where to place data?

Scikit-image (skimage) has a method called color.label2rgb which converts to an array of labels to a colored image. So you don't have to implement the coloring algorithm yourself.

According your code, how to use the Tensorboard ?

Hi,
I tried your code with the original carvana images, and it works. Then I prepare a custom dataset which has the same format as the carvana dataset, however, the training can't be converged. Can I send the dataset to you for testing?
Thanks!

请问我用自己的数据集运行data.py
报的这个错误是什么意思呢
Creating training images...
D:\anaconda\envs\tensorflow\lib\site-packages\keras_preprocessing\image\utils.py:107: UserWarning: grayscale is deprecated. Please use color_mode = "grayscale"
warnings.warn('grayscale is deprecated. Please use '
Done: 0/900 images
Done: 100/900 images
Done: 200/900 images
Done: 300/900 images
Done: 400/900 images
Done: 500/900 images
Done: 600/900 images
Done: 700/900 images
Done: 800/900 images
Traceback (most recent call last):
File "D:/Paper_project/unet-rgb-master/data.py", line 93, in
mydata.create_train_data()
File "D:/Paper_project/unet-rgb-master/data.py", line 41, in create_train_data
np.save(self.npy_path + '/imgs_train.npy', imgdatas)
File "<array_function internals>", line 6, in save
File "D:\anaconda\envs\tensorflow\lib\site-packages\numpy\lib\npyio.py", line 524, in save
file_ctx = open(file, "wb")
FileNotFoundError: [Errno 2] No such file or directory: './npydata/imgs_train.npy'
loading done

Process finished with exit code 1

Hi,

Great post and nicely coded. Your coding is very clear.

If we apply pre processing like k-means before semantic segmentation would it help to segment better ?

Traceback (most recent call last):
File "unet.py", line 137, in
model = myunet.get_unet()
File "unet.py", line 59, in get_unet
merge6 = merge([drop4, up6], mode='concat', concat_axis=3)
TypeError: 'module' object is not callable
can anyone help?

Hello,

Firstly thank you very much for your project. I have a problem about large dataset. I have 10 thousand images and masks.
I can create .npy files. When I run unet.py files with 10 thousand images I get error Memory Usage Exceeds. When I decrease the number of dataset (max 600 images) no any error. I think batch size don't work in model.fit. Do you have any advice or solution?

Best regards.

我按照博主给的代码训练模型，发生了MemoryError，可能是我的数据集比较大加上电脑内存不够，所以我现在想通过直接读取图片的方法来训练模型。请问博主能给一些处理数据的建议吗？