Upload README.md with huggingface_hub

README.md

@@ -28,7 +28,7 @@ This is a **paired regression** task trained with a noise-label objective (the g
 ## Dataset
 
 - **Source**: SEG C3 pre-stack synthetic data, 9 regular shot gathers
-- **Geometry**: 201 traces × 625 time samples per shot, dt = 8 ms
+- **Geometry**: 201 traces × 625 time samples per shot, dt = 2 ms
 - **Noise modeling**: Reflection signals modeled with the acoustic wave equation; ground roll modeled with the elastic wave equation to capture its dispersive, low-velocity character
 - **Split**: Shot-level (FFID) sequential 7:1:1 — 7 training shots, 1 validation, 1 held-out test
 
@@ -38,13 +38,13 @@ Five ground-roll intensity levels produce paired noisy / noise-label records:
 
 | Level | SNR (dB) | PSNR (dB) | SSIM | MAE | MSE | RMSE |
 |-------|----------|-----------|------|-----|-----|------|
-| 1.0   | 2.71     | 15.81     | 0.9480 | 0.030624 | 0.026232 | 0.161964 |
-| 3.0   | -6.83    | 6.27      | 0.9418 | 0.091871 | 0.236091 | 0.485892 |
-| 5.0   | -11.27   | 1.83      | 0.9402 | 0.153118 | 0.655809 | 0.809820 |
-| 7.0   | -14.19   | -1.09     | 0.9395 | 0.214366 | 1.285386 | 1.133749 |
-| 9.0   | -16.37   | -3.27     | 0.9390 | 0.275613 | 2.124822 | 1.457677 |
+| 1.0   | 2.71     | 21.83     | 0.95 | 0.020000 | 0.006558 | 0.080000 |
+| 3.0   | -6.83    | 18.31     | 0.95 | 0.020000 | 0.014756 | 0.120000 |
+| 5.0   | -11.27   | 17.40     | 0.95 | 0.030000 | 0.018217 | 0.130000 |
+| 7.0   | -14.19   | 16.97     | 0.95 | 0.030000 | 0.020084 | 0.140000 |
+| 9.0   | -16.37   | 16.73     | 0.95 | 0.030000 | 0.021248 | 0.150000 |
 
-*Metrics computed on the full dataset in the original amplitude domain before normalization.*
+*Metrics computed on the test set (2D flattened shot gathers) in the normalized domain before denoising.*
 
 ## Model Architectures
 
@@ -78,13 +78,11 @@ Each subdirectory corresponds to one experiment: a model architecture trained at
 
 | Hyperparameter | Value |
 |----------------|-------|
-| Loss | MSE (predicted noise vs. label noise) |
-| Optimizer | AdamW (lr=1e-3, weight_decay=1e-4) |
+| Loss | MSE (noise-prediction models) / GAN+L1 (pix2pix) / L1 (DDPM) / hybrid MSE+AFM (enhanced) |
+| Optimizer | Adam / AdamW (lr=1e-4–1e-3, varies per model) |
 | Scheduler | Cosine annealing (min_lr=1e-6) |
-| Epochs | 200 |
+| Epochs | 100–200 (varies per model) |
 | Gradient clipping | 1.0 (max norm) |
-| Batch size | 196 |
-| Distributed training | 2 × NVIDIA RTX 4090, DDP |
 | Seeds | 42, 43, 44 per experiment |
 
 ## Usage
@@ -113,17 +111,60 @@ state_dict = torch.load(ckpt_path, map_location="cpu", weights_only=True)
 
 See the companion benchmark documentation for detailed experimental setup and full evaluation results.
 
-## Results (SNR dB)
+## Results
 
-| Model | Params (M) | Level 1.0 | Level 3.0 | Level 5.0 | Level 7.0 | Level 9.0 |
-|-------|:----------:|:---------:|:---------:|:---------:|:---------:|:---------:|
-| Raw (noisy) | — | 2.71 | −6.83 | −11.27 | −14.19 | −16.37 |
-| UNet | 7.76 | 28.39±1.09 | 23.07±0.10 | 20.22±0.32 | 17.90±0.28 | 16.60±0.46 |
-| ResUNet | 8.11 | 31.62±0.56 | 22.65±0.93 | 18.11±2.12 | 16.60±1.44 | 14.61±0.01 |
-| DnCNN | 0.56 | 31.67±0.11 | 30.02±0.43 | 22.91±3.74 | — | — |
-| Attention UNet | 7.85 | 28.79±0.46 | 23.10±1.00 | 19.66±0.22 | 17.57±0.11 | 16.61±0.19 |
+Mean ± std over 3 seeds on the held-out test shot (FFID=9), evaluated on 2D-flattened data in the normalized domain. Raw (noisy) is the input before denoising.
+
+### Noise Level 1.0
+
+| Method | Params (M) | SNR (dB) | PSNR (dB) | SSIM | MAE | MSE | RMSE |
+| --- | :---: | :---: | :---: | :---: | :---: | :---: | :---: |
+| Raw (noisy) | — | 2.71±0.00 | 21.83±0.00 | 0.95±0.00 | 0.020000±0.00 | 0.006558±0.00 | 0.080000±0.00 |
+| UNet | 7.76 | 28.39±1.09 | 47.51±1.09 | 1.00±0.00 | 0.000000±0.00 | 0.000018±0.00 | 0.000000±0.00 |
+| ResUNet | 8.11 | 31.62±0.56 | 50.74±0.57 | 1.00±0.00 | 0.000000±0.00 | 0.000008±0.00 | 0.000000±0.00 |
+| DnCNN | 0.14 | 19.74±0.32 | 38.86±0.32 | 0.98±0.01 | 0.000000±0.00 | 0.000130±0.00 | 0.010000±0.00 |
+| Attention UNet | 7.85 | 29.48±1.35 | 48.60±1.35 | 1.00±0.00 | 0.000000±0.00 | 0.000014±0.00 | 0.000000±0.00 |
+
+### Noise Level 3.0
+
+| Method | Params (M) | SNR (dB) | PSNR (dB) | SSIM | MAE | MSE | RMSE |
+| --- | :---: | :---: | :---: | :---: | :---: | :---: | :---: |
+| Raw (noisy) | — | -6.83±0.00 | 18.31±0.00 | 0.95±0.00 | 0.020000±0.00 | 0.014756±0.00 | 0.120000±0.00 |
+| UNet | 7.76 | 23.07±0.10 | 48.21±0.10 | 1.00±0.00 | 0.000000±0.00 | 0.000015±0.00 | 0.000000±0.00 |
+| ResUNet | 8.11 | 22.65±0.93 | 47.79±0.93 | 0.99±0.00 | 0.000000±0.00 | 0.000017±0.00 | 0.000000±0.00 |
+| DnCNN | 0.14 | 17.24±0.23 | 42.38±0.23 | 0.99±0.00 | 0.000000±0.00 | 0.000058±0.00 | 0.010000±0.00 |
+| Attention UNet | 7.85 | 24.40±1.22 | 49.54±1.22 | 1.00±0.00 | 0.000000±0.00 | 0.000011±0.00 | 0.000000±0.00 |
+
+### Noise Level 5.0
+
+| Method | Params (M) | SNR (dB) | PSNR (dB) | SSIM | MAE | MSE | RMSE |
+| --- | :---: | :---: | :---: | :---: | :---: | :---: | :---: |
+| Raw (noisy) | — | -11.27±0.00 | 17.40±0.00 | 0.95±0.00 | 0.030000±0.00 | 0.018217±0.00 | 0.130000±0.00 |
+| UNet | 7.76 | 20.22±0.32 | 48.88±0.32 | 1.00±0.00 | 0.000000±0.00 | 0.000013±0.00 | 0.000000±0.00 |
+| ResUNet | 8.11 | 18.11±2.12 | 46.77±2.12 | 0.99±0.00 | 0.000000±0.00 | 0.000023±0.00 | 0.003333±0.01 |
+| DnCNN | 0.14 | 15.57±0.22 | 44.23±0.22 | 0.99±0.00 | 0.000000±0.00 | 0.000038±0.00 | 0.010000±0.00 |
+| Attention UNet | 7.85 | 21.96±1.06 | 50.62±1.06 | 1.00±0.00 | 0.000000±0.00 | 0.000009±0.00 | 0.000000±0.00 |
+
+### Noise Level 7.0
+
+| Method | Params (M) | SNR (dB) | PSNR (dB) | SSIM | MAE | MSE | RMSE |
+| --- | :---: | :---: | :---: | :---: | :---: | :---: | :---: |
+| Raw (noisy) | — | -14.19±0.00 | 16.97±0.00 | 0.95±0.00 | 0.030000±0.00 | 0.020084±0.00 | 0.140000±0.00 |
+| UNet | 7.76 | 17.90±0.28 | 49.06±0.28 | 1.00±0.00 | 0.000000±0.00 | 0.000012±0.00 | 0.000000±0.00 |
+| ResUNet | 8.11 | 16.60±1.44 | 47.76±1.44 | 0.99±0.00 | 0.000000±0.00 | 0.000017±0.00 | 0.000000±0.00 |
+| DnCNN | 0.14 | 15.18±0.13 | 46.34±0.13 | 0.99±0.00 | 0.000000±0.00 | 0.000023±0.00 | 0.000000±0.00 |
+| Attention UNet | 7.85 | 17.94±2.79 | 49.10±2.79 | 1.00±0.01 | 0.000000±0.00 | 0.000014±0.00 | 0.000000±0.00 |
+
+### Noise Level 9.0
+
+| Method | Params (M) | SNR (dB) | PSNR (dB) | SSIM | MAE | MSE | RMSE |
+| --- | :---: | :---: | :---: | :---: | :---: | :---: | :---: |
+| Raw (noisy) | — | -16.37±0.00 | 16.73±0.00 | 0.95±0.00 | 0.030000±0.00 | 0.021248±0.00 | 0.150000±0.00 |
+| UNet | 7.76 | 16.60±0.46 | 49.70±0.46 | 1.00±0.00 | 0.000000±0.00 | 0.000011±0.00 | 0.000000±0.00 |
+| ResUNet | 8.11 | 14.61±0.01 | 47.72±0.01 | 0.99±0.00 | 0.000000±0.00 | 0.000017±0.00 | 0.000000±0.00 |
+| DnCNN | 0.14 | 14.87±0.19 | 47.97±0.19 | 0.99±0.00 | 0.000000±0.00 | 0.000016±0.00 | 0.000000±0.00 |
+| Attention UNet | 7.85 | 15.79±2.76 | 48.88±2.76 | 0.99±0.01 | 0.000000±0.00 | 0.000015±0.00 | 0.000000±0.00 |
 
-Mean ± std over 3 seeds. All models achieve 15–31 dB SNR improvement. DnCNN delivers the best performance at low-to-mid noise levels with the smallest footprint (0.56 M parameters). See the benchmark documentation for per-level detailed metrics (PSNR, SSIM, MAE, MSE, RMSE).
 
 ## References