update train_cnn.py and shader

1 files changed, 43 insertions, 0 deletions

diff --git a/workspaces/main/shaders/cnn/cnn_conv5x5.wgsl b/workspaces/main/shaders/cnn/cnn_conv5x5.wgsl
index 5136740..0f261dd 100644
--- a/workspaces/main/shaders/cnn/cnn_conv5x5.wgsl
+++ b/workspaces/main/shaders/cnn/cnn_conv5x5.wgsl
@@ -83,3 +83,46 @@ fn cnn_conv5x5_7to1(
 
   return sum;  // Output in [-1,1]
 }
+
+// Source layer: 7→4 channels (RGBD output)
+// Normalizes [0,1] input to [-1,1] internally
+fn cnn_conv5x5_7to4_src(
+  tex: texture_2d<f32>,
+  samp: sampler,
+  uv: vec2<f32>,
+  resolution: vec2<f32>,
+  weights: array<array<f32, 8>, 100>
+) -> vec4<f32> {
+  let step = 1.0 / resolution;
+
+  let original = (textureSample(tex, samp, uv) - 0.5) * 2.0;
+  let gray = 0.2126*original.r + 0.7152*original.g + 0.0722*original.b;
+  let uv_norm = (uv - 0.5) * 2.0;
+
+  var sum = vec4<f32>(0.0);
+  var pos = 0;
+
+  for (var dy = -2; dy <= 2; dy++) {
+    for (var dx = -2; dx <= 2; dx++) {
+      let offset = vec2<f32>(f32(dx), f32(dy)) * step;
+      let rgbd = (textureSample(tex, samp, uv + offset) - 0.5) * 2.0;
+
+      let inputs = array<f32, 7>(
+        rgbd.r, rgbd.g, rgbd.b, rgbd.a,
+        uv_norm.x, uv_norm.y, gray
+      );
+
+      for (var out_c = 0; out_c < 4; out_c++) {
+        let idx = pos * 4 + out_c;
+        var channel_sum = weights[idx][7];
+        for (var in_c = 0; in_c < 7; in_c++) {
+          channel_sum += weights[idx][in_c] * inputs[in_c];
+        }
+        sum[out_c] += channel_sum;
+      }
+      pos++;
+    }
+  }
+
+  return sum;
+}