https://github.com/andreaazzini/segnet.tf

SegNet-like Autoencoders in TensorFlow
https://github.com/andreaazzini/segnet.tf

computer-vision convolutional-autoencoder deep-learning segmentation segnet tensorflow

Last synced: about 2 months ago
JSON representation

SegNet-like Autoencoders in TensorFlow

• Host: GitHub
• URL: https://github.com/andreaazzini/segnet.tf
• Owner: andreaazzini
• Created: 2016-11-29T16:53:17.000Z (almost 8 years ago)
• Default Branch: master
• Last Pushed: 2017-02-27T11:18:39.000Z (over 7 years ago)
• Last Synced: 2024-05-19T23:36:50.355Z (4 months ago)
• Topics: computer-vision, convolutional-autoencoder, deep-learning, segmentation, segnet, tensorflow
• Language: Python
• Homepage:
• Size: 15.9 MB
• Stars: 66
• Watchers: 5
• Forks: 26
• Open Issues: 7
• Metadata Files:
  • Readme: README.md

Awesome Lists containing this project

README

        
# SegNet

SegNet is a TensorFlow implementation of the [segmentation network proposed by Kendall et al.](http://mi.eng.cam.ac.uk/projects/segnet/), with cool features

like strided deconvolution, a minified architecture and more.

## Configuration

Create a `config.py` file, containing color maps, working dataset and other options.

```

autoencoder = 'segnet'

colors = {

  'segnet-32': [

    [64, 128, 64],   # Animal

    [192, 0, 128],   # Archway

    [0, 128, 192],   # Bicyclist

    [0, 128, 64],    # Bridge

    [128, 0, 0],     # Building

    [64, 0, 128],    # Car

    [64, 0, 192],    # CartLuggagePram

    [192, 128, 64],  # Child

    [192, 192, 128], # Column_Pole

    [64, 64, 128],   # Fence

    [128, 0, 192],   # LaneMkgsDriv

    [192, 0, 64],    # LaneMkgsNonDriv

    [128, 128, 64],  # Misc_Text

    [192, 0, 192],   # MotorcycleScooter

    [128, 64, 64],   # OtherMoving

    [64, 192, 128],  # ParkingBlock

    [64, 64, 0],     # Pedestrian

    [128, 64, 128],  # Road

    [128, 128, 192], # RoadShoulder

    [0, 0, 192],     # Sidewalk

    [192, 128, 128], # SignSymbol

    [128, 128, 128], # Sky

    [64, 128, 192],  # SUVPickupTruck

    [0, 0, 64],      # TrafficCone

    [0, 64, 64],     # TrafficLight

    [192, 64, 128],  # Train

    [128, 128, 0],   # Tree

    [192, 128, 192], # Truck_Bus

    [64, 0, 64],     # Tunnel

    [192, 192, 0],   # VegetationMisc

    [0, 0, 0],       # Void

    [64, 192, 0]     # Wall

  ]

}

gpu_memory_fraction = 0.7

strided = True

working_dataset = 'segnet-32'

```

Two kinds of architectures are supported at the moment: the original SegNet

Encoder-Decoder (`segnet`), and a smaller version of the same (`mini`), which

can be used for simpler segmentation problems. I suggest to use `strided = True`

for faster and more reliable results.

The `dataset_name` needs to match the data directories you create in your `input` folder. You can use `segnet-32` and `segnet-13` to replicate the aforementioned Kendall et al. experiments.

## Train and test

Generate your TFRecords using `tfrecorder.py`. In order to do so, put your PNG

images in a `raw` folder, as follows:

```

input/

    raw/

        train/

        train-labels/

        test/

        test-labels/

```

Once you have your TFRecords, train SegNet with `python src/train.py`. Analogously, test it with `python src/test.py`.