search

shaoyanpan / 2D-Medical-Denoising-Diffusion-Probabilistic-Model-

This is the repository for the paper "2D Medical Image Synthesis Using Transformer-based Denoising Diffusion Probabilistic Model".
https://iopscience.iop.org/article/10.1088/1361-6560/acca5c
MIT License
35 stars 9 forks source link

Evaluation Metric #16

Open ZhilingYan opened 4 months ago

ZhilingYan commented 4 months ago

Hi, this is an interesting work! Could you show more details and the code about your evaluation metrics? especially FDS and DS. Thanks.

shaoyanpan commented 4 months ago

Sorry I could not find the evaluation script anymore. But here are my implementations in MATLAB:

FDS: First, calculate TSNE features (let's say 2 features, you can select the number as your preference) from the synthetic image and true images (let's say we have 100 of each of them. "https://www.mathworks.com/help/stats/tsne.html" Now we have two matrices, each of them are 2x100. Then we calculate the KL between these two matrices: "https://www.mathworks.com/matlabcentral/fileexchange/20688-kullback-leibler-divergence".

DS: Use for loops to calculate the ssim for every synthetic image and all other synthetic images. For example, we have 3 synthetic images, the ssim matrix is 2x3. Each synthetic image has the neighbor with the highest ssim (so this pair is the most similar pair). Then we compute an average of all the highest ssim. The idea is that if even the most similar pair has low ssim, that mean the synthetic images are different.