Semantic_Segmentation

This repo shows a sample code to implement semantic segmenation (people segmentation) using unet + mobilenetv2. The dataset I used was coco2017.

Getting Started

Prerequisites

• Keras 2.4.3
• Tensorflow 2.2.0
• Coco API
• Opencv for python

Usage via command line

Training the Model

python unet_semantic_seg_train.py

Training Stats

The training set is coco/train2017 and the validation set is coco/val2017. The pretained model under model is getting 96.42% accuracy for training set and 92.9% accuracy for validation set.

Inference

python unet_semantic_seg_inference.py

The testing set used for inference is test2017. Some results are shown below:

References

• https://towardsdatascience.com/master-the-coco-dataset-for-semantic-image-segmentation-part-1-of-2-732712631047
• https://www.tensorflow.org/tutorials/images/segmentation
• https://cocodataset.org/#download