Now we have to figure out the structure of each chunk. Start with the "TX2D" chunk first. Jump to 0x1800, where we can see a lot of "TX2D" markers.
For convenience of analysis, we need to aligne them like this:

Let's see how many markers there are: 0x12C / 0x14(20) = 0xF. So the 00 00 00 0F at the header is the count of these markers. Same thing can be applied to
the "KFSQ" chunk. Therefore we get almost the entire header reversed.

Then we continue our research on the "TX2D" chunk. First we need to have an overall idea of how this chunk is organized:
the "TX2D" Table, a filename Table, then some more alignable data.
Let's try to have a look at the "TX2D" table and see if we can find something.

We notice that the values in each green frame are all in an increasing order. That means these values could be some offsets. And since they're all minner
than the current position, they should be relative to somewhere around.
Then what would these offsets direct to? Of course, the filename table, and the unknown data followed probably, what else?
With some assumptions and validation we soon can figure out the answers:

All these offsets are relative to where the "TX2D" table begins. While the structure of the texture info header table is like:

But the offset is relative to where the textures data chunk starts.