eragonruan/refinenet-image-segmentation

a tensorflow implement of refinenet ,RefineNet: Multi-Path Refinement Networks for High-Resolution Semantic Segmentation

refinenet-image-segmentation's Issues

advise for the code

The original code is written in matlab, and the network architectures are appended in the code package.
It's my honor to read your code in tensorflow, but there are some problems in your code different from the orignal network architectures. With respect, I list these codes in below pictures.



I will be waiting for your reply.

the result on pascal voc 2012 validation set

Hi , Thanks for sharing your coding .
I want to know your training result.
Can the model that you trained achieve iou score 83.4% on val set?
Thanks.

the utils.aug file

Hi,
I want to know the features of this function,I want to try.Can you tell some thing about it? Thank you very much!

Tensor not found Error

Hi @eragonruan ,
I am getting following error when i am running demo.py file.

tensorflow.python.framework.errors_impl.NotFoundError: Tensor name "feature_fusion/Conv/BatchNorm/gamma/ExponentialMovingAverage" not found in checkpoint files "resnet_v1_101.ckpt"

[[Node: save/RestoreV2_1 = RestoreV2[dtypes=[DT_FLOAT], _device="/job:localhost/replica:0/task:0/cpu:0"](_arg_save/Const_0_0, save/RestoreV2_1/tensor_names, save/RestoreV2_1/shape_and_slices)]]

Thanks & Regards,
Akansha Shah

error when run multi_gpu_train.py

when I run multi_gpu_train.py, it occurred that "tensorflow.python.framework.errors_impl.InvalidArgumentError: Input to reshape is a tensor with 786432 values, but the requested shape has 262144"
I don't know what caused this error, wish you can help me with it!
thanks!
@eragonruan

Weighted Cross Entropy loss function

Hi,
First thank you for this implementation.
I didn't test it yet but I'm about to, for custom data, a 3 classes segmentation task,
where implementing a weighted cross entropy function should be determinant.
What changes in the code should be done to do that ?
(I don't master tf yet, I worked on caffe before, a tip would be gratefuly taken)

Thanks in advance !

test.py problem

thanks for your code very much first！
can the test.py work now? and the miou can be obtained?

Error when run demo.py

'''NotFoundError (see above for traceback): Key feature_fusion/Conv_37/BatchNorm/beta/ExponentialMovingAverage not found in checkpoint'''
The ckpt is your trained model.

Thanks.

the color_map file

Hi,
Thank you very much for your help given me！I want to use my own data to train the model, but for color_map file I'm not particularly aware of how this is generated.Hope your reply.

Can you share your validation tfrecord?

Hi, @eragonruan .
Thanks for your sharing.
Can you sharing your validation set tfrecord?
In my view, the validation set does not need augmentation.
Thanks.
Have a good holiday

tensorflow version?

Why did I get a bad result?

I use mulit_gpu_train.py training data (the Tfrecord file you provided). Parameter ‘restore’ I chose False .When I choose true, the error is ‘'''NotFoundError (see above for traceback): Key feature_fusion/Conv_37/BatchNorm/beta/ExponentialMovingAverage not found in checkpoint'''’. so I chose False. And parameter ‘max_steps’ I chose 100000.
Training is done and the demo is running poorly.
why?

Unable to download model

The url in which you have uploaded the model is not working

chained residual pooling

Thanks for sharing
I think that at the chained residual pooling block there is a mistake.
Based on Fig. 3 of the paper, the input to the second, third etc. Pool layer is the output of the previous one, not the output of the ReLU layer:

def ChainedResidualPooling(inputs,features=256):
    net_relu=tf.nn.relu(inputs)
    net=slim.max_pool2d(net_relu, [5, 5],stride=1,padding='SAME')
    net=slim.conv2d(net,features,3)
    net_sum_1=tf.add(net,net_relu)

    # net = slim.max_pool2d(net_relu, [5, 5], stride=1, padding='SAME')
    net = slim.max_pool2d(net, [5, 5], stride=1, padding='SAME')

    net = slim.conv2d(net, features, 3)
    net_sum_2=tf.add(net,net_sum_1)

    return net_sum_2

can you achieve comparable result with this model?

Finished version

Hi,
Just wanted to know if you have a final version since you said it was a demo,
I'm wondering what's different from the version of the authors ...
Will try to implement it in Rstudio trying to use your code as an example,
I hope I'll succeed !
Thanks,
Joris

train other data and change the number of labels

Hi there,
Hope you are doing well.
I want to train other data using your code but I found changing the "num_classes" in mult_gpu_train.py is not working. Then I change the model function in model.py . I changed the "f_score" ,but the computer is out of memory.
What's wrong with my change? Can you give me help? Looking forward to your reply!

where is utils.aug

Under the utils folder, I didn't find the utils.aug file

your link to resnet weights is 404

I want to the version of tensorflow and whether I could use cpu to run the project

I want to the version of tensorflow and whether I could use cpu to run the project

I've met an error when runing 'python3 RefineNet/multi_gpu_train.py'

Traceback (most recent call last): File "RefineNet/multi_gpu_train.py", line 245, in <module> tf.app.run() File "/usr/local/lib/python3.5/dist-packages/tensorflow/python/platform/app.py", line 48, in run _sys.exit(main(_sys.argv[:1] + flags_passthrough)) File "RefineNet/multi_gpu_train.py", line 83, in main color_map = pickle.load(f) UnicodeDecodeError: 'ascii' codec can't decode byte 0xc0 in position 0: ordinal not in range(128)

Run demo.py error

I have the same problem proposed in previous issues.
When running demo.py, ERROR:
Key feature_fusion/Conv_37/BatchNorm/beta/ExponentialMovingAverage not found in checkpoint

I wonder if you save the checkpoint with ExponentialMovingAverage parameters ?

Unable to download the resnet_v1_101.ckpt model

Hi ,
I am intereted in your model,but the link which you gived of resnet_v1_101.ckpt is not working. Can you give me some help of this ?Thank you very much.