mally-2000 commited on
Commit
5162df9
·
verified ·
1 Parent(s): 4aad4b0

Focus README on model loading

Browse files

Prioritize from_pretrained usage, keep the clone demo secondary, and link to the full project code and paper.

Files changed (1) hide show
  1. README.md +34 -64
README.md CHANGED
@@ -12,71 +12,14 @@ tags:
12
 
13
  # Seismic-LDDPM
14
 
15
- Inference-only Hugging Face release for seismic impedance inversion. The model
16
- weights are hosted in this repository in Diffusers format:
17
-
18
- ```text
19
- vq_model/
20
- unet/
21
- scheduler/
22
- condition_encoder/
23
- ```
24
-
25
- The repository also includes a minimal Overthrust inference script. Training
26
- code and experiment configs are not included here.
27
-
28
- ## Quick Start
29
-
30
- ```bash
31
- git clone https://huggingface.co/mally-2000/seismic-lddpm
32
- cd seismic-lddpm
33
- pip install -r requirements.txt
34
- ```
35
-
36
- Run SAII-LDDPM:
37
-
38
- ```bash
39
- python infer.py
40
- ```
41
-
42
- Run SAII-CLDM:
43
-
44
- ```bash
45
- python infer.py CLDM
46
- ```
47
-
48
- `infer.py` builds the bundled Overthrust test sample, runs inference, and also
49
- runs the fixed Overthrust evaluation by default. Outputs are written under:
50
-
51
- ```text
52
- outputs/infer_LDDPM/
53
- outputs/infer_CLDM/
54
- ```
55
-
56
- ## Results
57
-
58
- Local Overthrust impedance-domain results:
59
-
60
- | Method | Steps | PSNR | SSIM | PCC | RRE |
61
- |---|---:|---:|---:|---:|---:|
62
- | SAII-LDDPM | 1000 | 33.4413 | 0.9554 | 0.9957 | 0.0324 |
63
- | SAII-CLDM | 30 | 33.1312 | 0.9494 | 0.9950 | 0.0342 |
64
 
65
- ## Python Loading
66
 
67
- Clone the repository and load local weights:
68
-
69
- ```python
70
- import torch
71
- from codes.pipeline import SeismicImpInvLDDPMPipeline
72
-
73
- pipe = SeismicImpInvLDDPMPipeline.from_pretrained(
74
- ".",
75
- torch_dtype=torch.float32,
76
- ).to("cuda")
77
- ```
78
-
79
- Or let Diffusers download the model weights and custom pipeline code:
80
 
81
  ```python
82
  import torch
@@ -90,7 +33,7 @@ pipe = DiffusionPipeline.from_pretrained(
90
  ).to("cuda")
91
  ```
92
 
93
- For SAII-CLDM remote loading:
94
 
95
  ```python
96
  pipe = DiffusionPipeline.from_pretrained(
@@ -100,3 +43,30 @@ pipe = DiffusionPipeline.from_pretrained(
100
  trust_remote_code=True,
101
  ).to("cuda")
102
  ```
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
12
 
13
  # Seismic-LDDPM
14
 
15
+ Open model weights for seismic impedance inversion in Diffusers format. This
16
+ repository is for inference and model loading; the full project code is at
17
+ [GeoAI-INV/SAII-CLDM](https://github.com/GeoAI-INV/SAII-CLDM), and the paper is
18
+ available on [arXiv](https://arxiv.org/html/2506.13529v1).
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
19
 
20
+ ## Load From Hugging Face
21
 
22
+ SAII-LDDPM:
 
 
 
 
 
 
 
 
 
 
 
 
23
 
24
  ```python
25
  import torch
 
33
  ).to("cuda")
34
  ```
35
 
36
+ SAII-CLDM:
37
 
38
  ```python
39
  pipe = DiffusionPipeline.from_pretrained(
 
43
  trust_remote_code=True,
44
  ).to("cuda")
45
  ```
46
+
47
+ The inference algorithms are implemented in `codes/pipeline.py`. The root
48
+ `pipeline.py` and `pipeline_cldm.py` files are thin entry points used by
49
+ Diffusers remote loading.
50
+
51
+ ## Run The Bundled Demo
52
+
53
+ ```bash
54
+ git clone https://huggingface.co/mally-2000/seismic-lddpm
55
+ cd seismic-lddpm
56
+ pip install -r requirements.txt
57
+
58
+ python infer.py # SAII-LDDPM
59
+ python infer.py CLDM # SAII-CLDM
60
+ ```
61
+
62
+ `infer.py` uses the bundled Overthrust sample and writes outputs under
63
+ `outputs/infer_LDDPM/` or `outputs/infer_CLDM/`.
64
+
65
+ ## Overthrust Results
66
+
67
+ Impedance-domain metrics on the bundled Overthrust setting:
68
+
69
+ | Method | Steps | PSNR | SSIM | PCC | RRE |
70
+ |---|---:|---:|---:|---:|---:|
71
+ | SAII-LDDPM | 1000 | 33.4413 | 0.9554 | 0.9957 | 0.0324 |
72
+ | SAII-CLDM | 30 | 33.1312 | 0.9494 | 0.9950 | 0.0342 |