search

tstandley / image2mass

http://proceedings.mlr.press/v78/standley17a/standley17a.pdf
38 stars 16 forks source link

thickness mask #9

Open HHLiufighting opened 9 months ago

HHLiufighting commented 9 months ago

Hi, I'm Liuhonghao.

Hello, I discovered an interesting phenomenon when obtaining masks with different picture thicknesses. Some objects are relatively large, but the maximum value in the mask is only 0.3. Some objects are not very thick but the value may be relatively large. These How are values predicted? What are the normalization standards?

Thank you very much in advance, Liuhonghao email: [liuhonghao_q@foxmail.com]

tstandley commented 9 months ago

Hi Liuhonghao,

I believe thickness values are normalized such that a thickness of 1 would represent a thickness equal to the entire horizontal or vertical image. For example, a sphere that barely touches the sides of the image should have a max thickness of 1.

Running through the training set, the maximum value I see for thickness is 0.81, with most objects being smaller, around the 0.3 you mentioned.