summaryrefslogtreecommitdiff
path: root/workspaces/main/shaders/cnn/cnn_layer.wgsl
blob: e629cb2e2fd4d025c9331c926c13cb995f181f59 (plain)
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
// CNN layer shader - uses modular convolution snippets
// Supports multi-pass rendering with residual connections
// DO NOT EDIT - Generated by train_cnn.py

@group(0) @binding(0) var smplr: sampler;
@group(0) @binding(1) var txt: texture_2d<f32>;

#include "common_uniforms"
#include "cnn_activation"
#include "cnn_conv3x3"
#include "cnn_weights_generated"

struct CNNLayerParams {
    layer_index: i32,
    blend_amount: f32,
    _pad: vec2<f32>,
};

@group(0) @binding(2) var<uniform> uniforms: CommonUniforms;
@group(0) @binding(3) var<uniform> params: CNNLayerParams;
@group(0) @binding(4) var original_input: texture_2d<f32>;

@vertex fn vs_main(@builtin(vertex_index) i: u32) -> @builtin(position) vec4<f32> {
    var pos = array<vec2<f32>, 3>(
        vec2<f32>(-1.0, -1.0), vec2<f32>(3.0, -1.0), vec2<f32>(-1.0, 3.0)
    );
    return vec4<f32>(pos[i], 0.0, 1.0);
}

@fragment fn fs_main(@builtin(position) p: vec4<f32>) -> @location(0) vec4<f32> {
    // Match PyTorch linspace
    let uv = (p.xy - 0.5) / (uniforms.resolution - 1.0);
    let original_raw = textureSample(original_input, smplr, uv);
    let original = (original_raw - 0.5) * 2.0;  // Normalize to [-1,1]
    let gray = (dot(original_raw.rgb, vec3<f32>(0.2126, 0.7152, 0.0722)) - 0.5) * 2.0;
    var result = vec4<f32>(0.0);

    // Layer 0: 7→4 (RGBD output, normalizes [0,1] input)
    if (params.layer_index == 0) {
        result = cnn_conv3x3_7to4_src(txt, smplr, uv, uniforms.resolution, weights_layer0);
        result = cnn_tanh(result);
    }
    else if (params.layer_index == 1) {
        result = cnn_conv3x3_7to4(txt, smplr, uv, uniforms.resolution, gray, weights_layer1);
        result = cnn_tanh(result);  // Keep in [-1,1]
    }
    else if (params.layer_index == 2) {
        let sum = cnn_conv3x3_7to1(txt, smplr, uv, uniforms.resolution, gray, weights_layer2);
        let gray_out = 1.0 / (1.0 + exp(-sum));  // Sigmoid activation
        result = vec4<f32>(gray_out, gray_out, gray_out, 1.0);
        return mix(original_raw, result, params.blend_amount); // [0,1]
    }
    return result;  // [-1,1]
}