I am trying to reconstruct position from depth in World Space (WS), but my code produce incorrect results.

So what I do:

1) Create depth map:

for each vertex do:

viewPos = POSITION_WS * matrixView;

Depth = viewPos.z / farClipPlane;

2) Get corners of camera view frustrum in WS (cornerRayWS)

3) Use far plane corners and reconstruct depth for (x, y) -> <0,1>

cornerRayWS_interp = lerp(lerp(cornerRayWS[0], cornerRayWS[1], x),

lerp(cornerRayWS[2], cornerRayWS[3], x),

y);

reconstructed_posWS = camera_pos_ws + Depth * cornerRayWS_interp;

But my results are incorrect. You can see input calc values for one camera position and view matrices in atached screen. From that I reconstruct position.

This, this, and this are good for reference. Have you looked at those articles?