Simplify model README

Keep the model card focused on inference, quick start commands, core results, and remote pipeline loading.

README.md

@@ -12,33 +12,20 @@ tags:
 
 # Seismic-LDDPM
 
-This Hugging Face repository is an inference-only release for seismic impedance
-inversion. The trained model weights are hosted directly in this repository in
-Diffusers format, together with a minimal inference entry point and a bundled
-Overthrust benchmark sample. Training scripts and experiment configs are not
-included in this model repository.
-
-The model takes a low-frequency impedance image (`dipin`) and a seismic record
-(`record`) as conditions, then predicts the impedance image.
-
-## Repository Contents
-
-- `vq_model/`, `unet/`, `scheduler/`, `condition_encoder/`: trained model
-  weights and component configs hosted on Hugging Face.
-- `infer.py`: single-sample inference entry point. Defaults to SAII-LDDPM;
-  pass `CLDM` to run SAII-CLDM sampling.
-- `pipeline.py`: thin Diffusers/Hugging Face custom pipeline entry point for
-  SAII-LDDPM.
-- `pipeline_cldm.py`: thin Diffusers/Hugging Face custom pipeline entry point
-  for SAII-CLDM.
-- `codes/pipeline.py`: `SeismicImpInvLDDPMPipeline` and
-  `SeismicImpInvCLDMPipeline`.
-- `codes/dataset.py`: lightweight Overthrust dataset utilities.
-- `codes/util.py`: differentiable Overthrust forward operator used by CLDM.
-- `codes/eval_overthrust.py`: fixed Overthrust evaluation script.
-- `data/Overthrust_trueimp.mat`: bundled Overthrust benchmark input.
-
-## Installation
+Inference-only Hugging Face release for seismic impedance inversion. The model
+weights are hosted in this repository in Diffusers format:
+
+```text
+vq_model/
+unet/
+scheduler/
+condition_encoder/
+```
+
+The repository also includes a minimal Overthrust inference script. Training
+code and experiment configs are not included here.
+
+## Quick Start
 
 ```bash
 git clone https://huggingface.co/mally-2000/seismic-lddpm
@@ -46,81 +33,38 @@ cd seismic-lddpm
 pip install -r requirements.txt
 ```
 
-After cloning, the model weights are already in the local repository
-subdirectories listed above. The provided scripts load from `.` by default when
-running inside the cloned repository. CUDA is used automatically when available.
-
-## Single-Sample Inference
-
-Run the default SAII-LDDPM sampler:
+Run SAII-LDDPM:
 
 ```bash
 python infer.py
 ```
 
-Run SAII-CLDM sampling:
+Run SAII-CLDM:
 
 ```bash
 python infer.py CLDM
 ```
 
-The script builds one default Overthrust sample internally and writes:
+`infer.py` builds the bundled Overthrust test sample, runs inference, and also
+runs the fixed Overthrust evaluation by default. Outputs are written under:
 
 ```text
 outputs/infer_LDDPM/
 outputs/infer_CLDM/
 ```
 
-Each output directory contains `prediction.npy`, `target.npy`, and
-`comparison.png`.
-
-## Overthrust Evaluation
-
-Evaluate SAII-LDDPM on the bundled six-patch Overthrust benchmark:
-
-```bash
-python codes/eval_overthrust.py LDDPM --model . --output outputs/overthrust_LDDPM
-```
-
-Evaluate SAII-CLDM:
-
-```bash
-python codes/eval_overthrust.py CLDM --model . --output outputs/overthrust_CLDM
-```
-
-The evaluation script uses the fixed benchmark setup:
-
-- patch indices: `[0, 1, 2, 3, 4, 5]`
-- patch size: `256 x 256`
-- noise SNR: `15 dB`
-- low-frequency input: `dipin_v=0.012`
-- Ricker wavelet: `f0=30`, `phase=0`
-- seed: `1234`
-
-Outputs include stitched predictions, a comparison image, and
-`metrics_summary.json`.
-
-## Local Benchmark Results
-
-SAII-LDDPM, 1000 DDPM steps:
-
-| Space | PSNR | SSIM | PCC | RRE | NMSE |
-|---|---:|---:|---:|---:|---:|
-| Normalized | 30.7698 | 0.9339 | 0.9963 | 0.0435 | 0.001894 |
-| Impedance | 33.4413 | 0.9554 | 0.9957 | 0.0324 | 0.001050 |
-| VQ reconstruction | 37.7954 | 0.9677 | 0.9983 | 0.0209 | 0.000435 |
+## Results
 
-SAII-CLDM, 30 DDIM steps with model-driven resampling:
+Local Overthrust impedance-domain results:
 
-| Space | PSNR | SSIM | PCC | RRE | NMSE |
+| Method | Steps | PSNR | SSIM | PCC | RRE |
 |---|---:|---:|---:|---:|---:|
-| Normalized | 31.1638 | 0.9246 | 0.9956 | 0.0422 | 0.001778 |
-| Impedance | 33.1312 | 0.9494 | 0.9950 | 0.0342 | 0.001171 |
-| VQ reconstruction | 37.7954 | 0.9677 | 0.9983 | 0.0209 | 0.000435 |
+| SAII-LDDPM | 1000 | 33.4413 | 0.9554 | 0.9957 | 0.0324 |
+| SAII-CLDM | 30 | 33.1312 | 0.9494 | 0.9950 | 0.0342 |
 
-## Python Usage
+## Python Loading
 
-If you cloned this repository, import the pipeline class directly:
+Clone the repository and load local weights:
 
 ```python
 import torch
@@ -129,21 +73,10 @@ from codes.pipeline import SeismicImpInvLDDPMPipeline
 pipe = SeismicImpInvLDDPMPipeline.from_pretrained(
     ".",
     torch_dtype=torch.float32,
-    trust_remote_code=True,
 ).to("cuda")
-
-result = pipe(
-    dipin=dipin,
-    record=record,
-    num_inference_steps=1000,
-    seed=1234,
-)
-
-prediction = result.impedance_samples
 ```
 
-If you want Diffusers to download the model weights and custom pipeline code
-from Hugging Face automatically for SAII-LDDPM:
+Or let Diffusers download the model weights and custom pipeline code:
 
 ```python
 import torch
@@ -157,12 +90,9 @@ pipe = DiffusionPipeline.from_pretrained(
 ).to("cuda")
 ```
 
-For SAII-CLDM remote loading, use the CLDM custom pipeline file:
+For SAII-CLDM remote loading:
 
 ```python
-import torch
-from diffusers import DiffusionPipeline
-
 pipe = DiffusionPipeline.from_pretrained(
     "mally-2000/seismic-lddpm",
     custom_pipeline="pipeline_cldm",
@@ -170,6 +100,3 @@ pipe = DiffusionPipeline.from_pretrained(
     trust_remote_code=True,
 ).to("cuda")
 ```
-
-For CLDM, use `SeismicImpInvCLDMPipeline` and pass a seismic forward operator
-through `operator=...`; see `infer.py` for the minimal runnable example.