venkanna37 / label-pixels Goto Github PK

environment.yml file crated using anaconda commands. Because many packages installing with other packages, this file has to create again.

Create and add the link to readme.

Previously, there were three FCN architecture added in tools/models folder such as UNet, SegNet and ResUNet. To test the tool and reduce the size of the weights file, the number of convolutional layers reduced in the UNet architecture. Now, UNet with less convolutional layers renamed to unet_mini and the original UNet architecture is unet.

The architecture that used in examples is unet_mini, but example commands not updated.

Image data augmentation is a technique to increase the size of the training dataset. This is useful when there is less training data to train.

Currently, patch size needs to be added.

Using TensorFlow backend.
Traceback (most recent call last):
File "train.py", line 63, in
train(args)
File "train.py", line 20, in train
model = lu.select_model(args)
File "/home/user/Downloads/Label-Pixels/tools/models/lp_utils.py", line 23, in select_model
model = segnet_model.create_segnet(args)
File "/home/user/Downloads/Label-Pixels/tools/models/segnet_model.py", line 158, in create_segnet
encoder = VGG16_encoder(input_shape, init=True)
File "/home/user/Downloads/Label-Pixels/tools/models/segnet_model.py", line 80, in VGG16_encoder
weights_fname = write_new_VGG_weights(input_shape, weights_fname, outfname)
File "/home/user/Downloads/Label-Pixels/tools/models/segnet_model.py", line 67, in write_new_VGG_weights
del f["block1_conv1"]["block1_conv1_W_1:0"]
File "h5py/_objects.pyx", line 54, in h5py._objects.with_phil.wrapper
File "h5py/_objects.pyx", line 55, in h5py._objects.with_phil.wrapper
File "/home/user/anaconda3/envs/label-pixels/lib/python3.6/site-packages/h5py/_hl/group.py", line 264, in getitem
oid = h5o.open(self.id, self._e(name), lapl=self._lapl)
File "h5py/_objects.pyx", line 54, in h5py._objects.with_phil.wrapper
File "h5py/_objects.pyx", line 55, in h5py._objects.with_phil.wrapper
File "h5py/h5o.pyx", line 190, in h5py.h5o.open
KeyError: "Unable to open object (object 'block1_conv1' doesn't exist)"

really amazing paper. Is it convenient to provide me with some pre-trained models? My computer does not have a powerful GPU for training，but prediction should be ok. very thankful!!

Hi @venkanna37 ,

You have done a fantastic job, it would be great if you can add your thesis/journal in the readme section.

Regards,

Currently, the dataset has to be split manually before creating patches or before training the model. Adding a script or feature in patch_gen.py avoids manual splitting. Otherway is to create patches and CSV files for the entire dataset and then rows in the CSV file can be split into three sets (training, testing and validation).