https://github.com/oliver-batchelor/segmentation

Image segmentation using pytorch.
https://github.com/oliver-batchelor/segmentation

convolutional-neural-networks machinine-learning pytorch segmentation

Last synced: about 1 month ago
JSON representation

Image segmentation using pytorch.

• Host: GitHub
• URL: https://github.com/oliver-batchelor/segmentation
• Owner: oliver-batchelor
• Created: 2017-04-12T06:05:22.000Z (almost 8 years ago)
• Default Branch: master
• Last Pushed: 2018-05-16T03:41:16.000Z (over 6 years ago)
• Last Synced: 2023-03-24T04:57:57.188Z (almost 2 years ago)
• Topics: convolutional-neural-networks, machinine-learning, pytorch, segmentation
• Language: Python
• Homepage:
• Size: 762 KB
• Stars: 4
• Watchers: 2
• Forks: 0
• Open Issues: 0
• Metadata Files:
  • Readme: README.md

Awesome Lists containing this project

README

        
# Segmentation dataset

Make sure to checkout with --recurse-submodules, there's a submodule 'tools' for common bits and pieces)

It loads a folder full of images and masks (pixel value corresponds to class) with a config file.

An example dataset to train on can be found at:

https://drive.google.com/file/d/0B_mOCEqr7usZa2hra0xZUTJscE0/view

Or classes/images from the COCO/Pascal VOC dataset can be imported from scripts in the import/ folder.

# View the training or testing set and mask annotations:

  `python -m dataset.view --input /path/to/dataset --train (or --test)`

  Useful to check the preprocessing of images.

# View a mask file

  `python view_labels.py some/file.jpg.mask`

# Train a model:

  `python main.py --lr 0.1 --batch_size 4 --input /path/to/dataset --model "unet --depth 5" --epoch_size 1024`

## Common options:

  `--load`, load from a previous checkpoint and cointunue training  

  `--model`, specify model and model parameters (use quotes)  

  `--show`, show results of evaluating the model in training (sanity check)

# Evaluate a model on new image(s):

  `python test.py --batch /path/to/images --model log/model.pth --save results_path` (and/or `--show`)  

  `python test.py --image /my/image.jpg --model log/model.pth --show`