I'm having a problem with figuring out spaces during SSAO calculation. What i'm trying to achieve is basic ssao effectwith depth reconstruction (before diving in to more complicated normals reconstructions).

As far as i understand, most tutorial teach how to compute AO in view/eye space, which lead me to this code that recreates View Space Positions from depth:

vec3 viewSpacePosFromDepth(vec2 uv)
{
    float depth = texture(depthTexture, uv).r; //Depth texture is standard Z/W depth
    vec4 clipSpacePosition = vec4(uv * 2.0 - 1.0, 2.0 * depth - 1.0, 1.0);
    vec4 position = InverseProjection * clipSpacePosition;

    return position.xyz / position.w;
}

Results looks correct (here, slightly scaled):

However, when i try to apply SSAO, result is rotating with camera: https://streamable.com/twarwy​​ (also note different result on left and right of a screen, with similar geometry)

Code itself is rather simple:

void main()
{
    vec3 pos = viewSpacePosFromDepth(TexCoord); // View space position
    float ao = 0.0;

    for (int i = 0; i < KERNEL_SIZE; i++) 
    {
        vec3 samplePos = pos + Samples[i]; // Samples are random points in [-1, 1] space, so we're still in View Space
        vec4 offset = vec4(samplePos, 1.0);
        offset = Projection * offset; // Moving back to clip space
        offset.xy /= offset.w;
        offset.xy = offset.xy * 0.5 + vec2(0.5);

        float sampleDepth = viewSpacePosFromDepth(offset.xy).z; // again, sampleDepth is in View space

        if (abs(pos.z - sampleDepth) < radius) 
        {
            ao += step(sampleDepth, samplePos.z);
        }
    }

    ao = 1.0 - AO / KERNEL_SIZE;
    fragColor = vec4(pow(ao, 2.0));
} 

I'm not following, why the result is rotating with camera, when i'm doing all the calculations in View Space. Am i missing something?