cnncnnzh / tem_segmentation

A TEM Image Segmentation Project Featuring Convolutional Neuron Network with U-Net Architecture

U-Net for Segmentation of Transmission Electron Microscopy Images

Dataset

This project utilizes the TEM-ImageNet-v1.3 dataset, which can be downloaded from the following GitHub repository: https://github.com/xinhuolin/TEM-ImageNet-v1.3. The dataset includes images and circular masks, which are used in the training process of our model.

Usage

To run the training script train.py, you need to specify the directories of the images and circular masks from the TEM-ImageNet-v1.3 dataset. You can set these directories as arguments for dir_img and dir_mask in the train.py script.

The dir_checkpoint argument in the train.py script specifies the directory where the training checkpoints will be saved.

In the models directory, you can find five different models: unet_2_layer, unet_3_layer, unet_4_layer, unet_cnn, and unet_wo_skip. You can choose a specific model to train by inputting the corresponding number when prompted during the execution of train.py.

Please refer to the environment.yml file for the required package versions for this project.

References

TEM-ImageNet-v1.3 dataset: https://github.com/xinhuolin/TEM-ImageNet-v1.3

Sample results