hellochick / indoor-segmentation Goto Github PK

Can you provide the way to run the model on video feed?

Hi, Thank you for the code, it's really helpful.

I'm having problems trying to run the network in realtime. I'm new to tensorflow and I'm used to defining the graph outside the session then using a feeddict in a Session().run for the input placeholders.

how can I feed another input image after the first prediction had run?

Thanks,
Ramy Mounir

Is there any way to sharpen the corners of the objects in the output image? Currently, they appear rounded and approximated, which I suspect may be a result of preprocessing. I need the objects to have clearly defined corners for my application.

Hi

Can you please share the code for relabelling data? Thanks.

Im getting the following error while trying to run the code

File "inference.py", line 128, in <module>
    main()
  File "inference.py", line 85, in main
    net = DeepLabResNetModel({'data': tf.expand_dims(img, dim=0)}, is_training=False, num_classes=NUM_CLASSES)
  File "/Users/Downloads/Indoor-segmentation-master/network.py", line 48, in __init__
    self.setup(is_training, num_classes)
  File "/Users/Downloads/Indoor-segmentation-master/model.py", line 19, in setup
    (self.feed('data')
  File "/Users/Downloads/Indoor-segmentation-master/network.py", line 78, in feed
    if isinstance(fed_layer, basestring):
NameError: name 'basestring' is not defined

Hi.
I ran the inference using the command:
python inference.py --img_path=test_img.png --restore_from=/restore_weights/

but the error came out:

File "/home/nhan/git_workspace/Indoor-segmentation/network.py", line 78, in feed
if isinstance(fed_layer, basestring):
NameError: name 'basestring' is not defined

I look for basestring in network.py and basestring is really not defined.
Could you help me, please?

Thx for your work, but I can't download anything from google drive, would you please just fix it?

I previously change basestring to str and then tried running inference with
python inference.py --img_path="./input/image1.png" --restore_from="./restore_weights"

Any idea to resolve this?

When I run the inference.py using the models form the link of google drive, I get bad results.
this is how i run it:
python inference.py --img_path=input/IMG_0416_640x480.png --restore_from=ResNet50/model.ckpt-22700

Can you share the conda dependency versions.

Hi,

I was wondering if you could provide the model code for ResNet50 and/or for ICNet as I am receiving errors because of some layers not being a part of the checkpoint file. The prediction works very well for ResNet101 and I was wondering if I could make a comparison between the speed/accuracy tradeoff between these three networks. Thanks in advance!

Thanks for your contributions.
I want to try transfer learning with new datasets. Would you upload the train.py on this repo or send me the file?

Your work seems really helpful to me, but I'm having trouble accessing the pretrained model from Google Drive. Could you please help me fix this?

When I run the inference script with windows paths I get FileNotFoundError.

Input:
python inference.py --img_path input\IMG_0416_640x480.png --restore_from=pretrained_models\ResNet101

Error:
FileNotFoundError: [Errno 2] No such file or directory: './output/input\\IMG_0416_640x480.png'