Skip to content

Latest commit

 

History

History
47 lines (33 loc) · 2.89 KB

Dataset Prepare.md

File metadata and controls

47 lines (33 loc) · 2.89 KB
Raw

Dataset Preparation

In order to be able to start the training of CenterSAM, a setup of the dataset is required. Currently, we provide the example of using TissueNet dataset as the the target dataset in this github code, steps as follow:

Download and locate the dataset

For The original TissueNet dataset, you may download it by visit the official dataset website (A github account is required for browsing the data).

The TissueNet dataset is originally provided in format of three npz file, splited by train test and val purpose. Images and annotation information are all compose inside those npz file.

While the input of CenterSAM model need to seperately handle the images and annotations, you need to follow the Readme.md file inside the downloaded tissuenet_v1.1.zip file to manually extract the images and annotations. In order to provide a more convenient way to start training the model and give you better understand of dataset setup requiments, we provide extracted images and annotations in This Google Drive Link. Please download the images.zip and annotations.zip file via the link and locate as follow:

CenterSAM/
└── data/
    └── TissueNet/
        ├── images/
        │   ├── train1.png
        │   ├── train2.png
        │   ├── ...
        │   ├── test1.png        
        │   ├── ...
        │   ├── val1.png
        │   └── ...
        └── annotations/
            ├── train.json
            ├── val.json
            └── test.json

Now you can follow Quick Start to start the training process of CenterSAM.

In case of train custom dataset, please read the section below.

Train custom dataset

For a custom dataset, in addition to separating images and annotations and placing them under CenterSAM/data/{Your_dataset_name}/ as described above, you need to create a py file with the same name as your dataset in the CenterSAM/src/lib/datasets/dataset/ folder, Assuming that the name of your dataset is CUSTOM, you can perform the following steps to complete the configuration：

  • Copy TissueNet.py file under the CenterSAM/src/lib/datasets/dataset/, rename it to CUSTOM.py
  • Rename all TissueNet inside the file to CUSTOM
  • Replace the mean and std variable with the mean and standard deviation of your CUSTOM dataset images.
  • Modify the CenterSAM/src/lib/datasets/sample/dataset_factory.py file:
    • Import the CUSTOM.py file in the same way as TissueNet: add line from .dataset.CUSTOM import CUSTOM
    • Add 'TissueNet': TissueNet key-value paris in dataset_factory dictionary