minna-ust / Semantic-Segmentation-with-Unet

Pytorch implementation of Human(person) segmentation from RGB images.

Semantic-Segmentation-with-Unet

Pytorch implementation of Human(person) segmentation from RGB images.

Dataset

Dataset is available in https://supervise.ly/explore/projects/supervisely-person-dataset-23304/datasets Please make sure to arrange the Dataset tree as follows. dataset_dir(str) : path to the dataset(root dir) and arranged as follows.

├── Dataset
│   ├── sample.png
│      ├──images
│        ├── sample.png
│      ├── masks
│        ├── id[0].png
         └── id[i].png

Dependancies

1. Python3
2. Pytorch 1.1.0

Unet Model Evaluation.

jupyter notebook(Unet_Evaluation.ipynb) for data loader and model prediction is provided.

Usage

python train.py
If you want help - python train.py --help

UNET + Background Edit

A new feature has added with the existing Segmentation Model. We use the UNET model to transfer the input image colorful background to Grey . Since our model is trained only for person classes, demos are limited to input images with at least one person. I have provided a demo notebook to guide you to achieve this. Here are some demo samples. .

Before you start, pleae put the pretrained model in the main folder.

You can find the pretrained model here : https://drive.google.com/file/d/1iaJA5AvNmAuFUv1HbG0_gWVZWA-ZN4La/view?usp=sharing