H2-conv (3.17) step 14000: eval tables

image_video_conv/step14000/README.md

@@ -9,3 +9,17 @@ after the splat, splat_direct, 8-head mid attention, conv_low_in). Trained on im
 - `s14000_*.md`: eval tables (200 ImageNet val images; 20 OpenVid held-out clips, stride 8 and 2).
 - `launcher.sh`: the training flags (paths removed).
 
+
+Eval (200 ImageNet val images; 20 OpenVid clips):
+- s14000_img:
+    | 200 | 29.76 | 0.858 | 0.092 | 21.77 | 16420 |
+- s14000ema_img:
+    | 200 | 30.06 | 0.865 | 0.086 | 21.13 | 16420 |
+- s14000_st8:
+    | anchor | 32.37 | 0.919 | 0.070 | - |
+    | target, full state | 27.28 | 0.829 | 0.140 | 28626 |
+    | target, L = 0 | 27.17 | 0.828 | 0.144 | 24882 |
+- s14000_st2:
+    | anchor | 32.45 | 0.920 | 0.070 | - |
+    | target, full state | 29.40 | 0.879 | 0.103 | 28626 |
+    | target, L = 0 | 29.31 | 0.878 | 0.105 | 24882 |

image_video_conv/step14000/tables/s14000_img.md

@@ -0,0 +1,9 @@
+# Grain-Prior C-D v1 eval, step 14000
+
+query_init=cond geom_bias=True memory=True image ledger=16420 full state=28626
+
+## Images
+
+| n | PSNR | SSIM | LPIPS | rFID | numbers |
+|---|---|---|---|---|---|
+| 200 | 29.76 | 0.858 | 0.092 | 21.77 | 16420 |

image_video_conv/step14000/tables/s14000_st2.md

@@ -0,0 +1,21 @@
+# Grain-Prior C-D v1 eval, step 14000
+
+query_init=cond geom_bias=True memory=True image ledger=16420 full state=28626
+
+## video manifest (20 clips, payload 28626)
+
+| decode | PSNR | SSIM | LPIPS | numbers |
+|---|---|---|---|---|
+| anchor | 32.45 | 0.920 | 0.070 | - |
+| target, full state | 29.40 | 0.879 | 0.103 | 28626 |
+| target, L = 0 | 29.31 | 0.878 | 0.105 | 24882 |
+| target, L = 52 | 29.40 | 0.879 | 0.103 | 28626 |
+| novel view: grains + camera only (z = c_0, L = 0) | 21.81 | 0.605 | 0.243 | 24882 |
+| copy anchor (no model) | 24.81 | 0.752 | 0.115 | - |
+| copy last frame (no model) | 27.56 | 0.831 | 0.070 | - |
+
+target by record (full state), PSNR: 30.10 / 29.47 / 28.62
+
+gates: L0 - copy-last = 1.75 dB; novel view - copy anchor = -3.00 dB
+
+init_cos=0.705 base_cos_pass0=0.808 base_cos_pass1=0.815 base_cos_final=0.839 feat_cos=0.844 support_frac=1.000 psnr_base=29.307 xi_rot_deg=0.680 xi_trans=0.341 fov_deg=62.933 rho_std=0.019 mem_delta=0.033 sigma_mean=0.802 psnr_copy_last=27.560 splat_dropped_frac=0.000

image_video_conv/step14000/tables/s14000_st8.md

@@ -0,0 +1,21 @@
+# Grain-Prior C-D v1 eval, step 14000
+
+query_init=cond geom_bias=True memory=True image ledger=16420 full state=28626
+
+## video manifest (20 clips, payload 28626)
+
+| decode | PSNR | SSIM | LPIPS | numbers |
+|---|---|---|---|---|
+| anchor | 32.37 | 0.919 | 0.070 | - |
+| target, full state | 27.28 | 0.829 | 0.140 | 28626 |
+| target, L = 0 | 27.17 | 0.828 | 0.144 | 24882 |
+| target, L = 52 | 27.28 | 0.829 | 0.140 | 28626 |
+| novel view: grains + camera only (z = c_0, L = 0) | 17.49 | 0.392 | 0.383 | 24882 |
+| copy anchor (no model) | 19.51 | 0.573 | 0.221 | - |
+| copy last frame (no model) | 21.73 | 0.658 | 0.157 | - |
+
+target by record (full state), PSNR: 28.24 / 27.27 / 26.32
+
+gates: L0 - copy-last = 5.45 dB; novel view - copy anchor = -2.02 dB
+
+init_cos=0.701 base_cos_pass0=0.794 base_cos_pass1=0.803 base_cos_final=0.831 feat_cos=0.836 support_frac=1.000 psnr_base=27.177 xi_rot_deg=0.838 xi_trans=0.465 fov_deg=62.841 rho_std=0.020 mem_delta=0.038 sigma_mean=0.800 psnr_copy_last=21.728 splat_dropped_frac=0.004

image_video_conv/step14000/tables/s14000ema_img.md

@@ -0,0 +1,9 @@
+# Grain-Prior C-D v1 eval, step 14000
+
+query_init=cond geom_bias=True memory=True image ledger=16420 full state=28626
+
+## Images
+
+| n | PSNR | SSIM | LPIPS | rFID | numbers |
+|---|---|---|---|---|---|
+| 200 | 30.06 | 0.865 | 0.086 | 21.13 | 16420 |