Floating Point Compression Algorithm

Refer to IEEE floating point format description
Float numbers between 0 and 1 can be compressed.

Obviously, in the range of 0 to 1, a single-precision floating point number, index and symbol bit occupy 9 bits
The nine bits are not needed. We can delete them and only keep the 23bit of the decimal part to compress them.
You can compress a floating point number from 32-bit to 4-bit to 3-byte.

This is also a method for compressing and storing floating point numbers in tools such as 3dmax.
For example, in the unit-based Normal Vector, each floating point number is data in the range of 0 and 1.
Normally, it takes 12 bytes to represent the unitized vector in 3D space.
After this compression, only 9 bytes are required.