NCJ commited on
Commit
5473044
·
verified ·
1 Parent(s): 2a5582b

Document direct component loading

Browse files
Files changed (1) hide show
  1. README.md +17 -34
README.md CHANGED
@@ -10,14 +10,15 @@ tags:
10
 
11
  # RenderFormer V2
12
 
13
- > **Status:** This is a private pre-release bundle. Do not redistribute it or
14
- > depend on its layout until the release is approved.
15
-
16
  This repository bundles the two RenderFormer V2 renderer checkpoints and the
17
  material-latent models used by the data pipeline. Source code, installation
18
  instructions, scene conversion, and inference examples live in
19
  [RenderFormer Studio](https://github.com/iamNCJ/RenderFormer-Studio).
20
 
 
 
 
 
21
  ## Components
22
 
23
  | Path | Class | Resolution | Purpose |
@@ -35,37 +36,27 @@ external dependency and are not duplicated in this bundle.
35
 
36
  ## Load a renderer
37
 
38
- The current package accepts a local component directory. Download only the
39
- resolution you need, then construct the shared V2 pipeline:
40
 
41
  ```python
42
- from pathlib import Path
43
-
44
- import torch
45
- from huggingface_hub import snapshot_download
46
  from renderformer import RenderFormerV2Pipeline
47
 
48
- snapshot = Path(snapshot_download(
49
- "RenderFormer/renderformer-v2",
50
- allow_patterns=["transformer_512/*"],
51
- ))
52
  pipeline = RenderFormerV2Pipeline.from_pretrained(
53
- snapshot / "transformer_512",
 
54
  device="cuda",
55
- torch_dtype=torch.float16,
56
  )
 
57
  ```
58
 
59
- Use `transformer_2048` in both places for native-2048 inference. The CLI and
60
- pipeline documentation in RenderFormer Studio describe H5 preparation,
61
- precision, view batching, and tone mapping.
 
62
 
63
  ## Load material components
64
 
65
  ```python
66
- from pathlib import Path
67
-
68
- from huggingface_hub import snapshot_download
69
  from renderformer.models.material import (
70
  DiffuseSpecularToLatent,
71
  MaterialAutoencoder,
@@ -73,26 +64,18 @@ from renderformer.models.material import (
73
  PrincipledBRDFToLatentWithTransmission,
74
  )
75
 
76
- snapshot = Path(snapshot_download(
77
- "RenderFormer/renderformer-v2",
78
- allow_patterns=[
79
- "material_autoencoder/*",
80
- "diffspec_mapper/*",
81
- "metallic_mapper/*",
82
- "metallic_transmission_mapper/*",
83
- ],
84
- ))
85
  material_autoencoder = MaterialAutoencoder.from_pretrained(
86
- snapshot / "material_autoencoder", strict=True
87
  ).eval()
88
  diffspec_mapper = DiffuseSpecularToLatent.from_pretrained(
89
- snapshot / "diffspec_mapper", strict=True
90
  ).eval()
91
  metallic_mapper = PrincipledBRDFToLatent.from_pretrained(
92
- snapshot / "metallic_mapper", strict=True
93
  ).eval()
94
  transmission_mapper = PrincipledBRDFToLatentWithTransmission.from_pretrained(
95
- snapshot / "metallic_transmission_mapper", strict=True
96
  ).eval()
97
  ```
98
 
 
10
 
11
  # RenderFormer V2
12
 
 
 
 
13
  This repository bundles the two RenderFormer V2 renderer checkpoints and the
14
  material-latent models used by the data pipeline. Source code, installation
15
  instructions, scene conversion, and inference examples live in
16
  [RenderFormer Studio](https://github.com/iamNCJ/RenderFormer-Studio).
17
 
18
+ If this repository is access-protected during release staging, authenticate
19
+ with Hugging Face before loading it. One repository revision selects every
20
+ component below as an atomic, compatible set.
21
+
22
  ## Components
23
 
24
  | Path | Class | Resolution | Purpose |
 
36
 
37
  ## Load a renderer
38
 
39
+ `resolution` automatically selects `transformer_512` or `transformer_2048`:
 
40
 
41
  ```python
 
 
 
 
42
  from renderformer import RenderFormerV2Pipeline
43
 
 
 
 
 
44
  pipeline = RenderFormerV2Pipeline.from_pretrained(
45
+ "RenderFormer/renderformer-v2",
46
+ resolution=512,
47
  device="cuda",
 
48
  )
49
+ result = pipeline("scene.h5", resolution=512, precision="fp16")
50
  ```
51
 
52
+ Advanced callers may select a component explicitly with
53
+ `subfolder="transformer_512"`. The CLI and pipeline documentation in
54
+ RenderFormer Studio cover H5 preparation, precision, view batching, and tone
55
+ mapping.
56
 
57
  ## Load material components
58
 
59
  ```python
 
 
 
60
  from renderformer.models.material import (
61
  DiffuseSpecularToLatent,
62
  MaterialAutoencoder,
 
64
  PrincipledBRDFToLatentWithTransmission,
65
  )
66
 
67
+ repo = "RenderFormer/renderformer-v2"
 
 
 
 
 
 
 
 
68
  material_autoencoder = MaterialAutoencoder.from_pretrained(
69
+ repo, subfolder="material_autoencoder", strict=True
70
  ).eval()
71
  diffspec_mapper = DiffuseSpecularToLatent.from_pretrained(
72
+ repo, subfolder="diffspec_mapper", strict=True
73
  ).eval()
74
  metallic_mapper = PrincipledBRDFToLatent.from_pretrained(
75
+ repo, subfolder="metallic_mapper", strict=True
76
  ).eval()
77
  transmission_mapper = PrincipledBRDFToLatentWithTransmission.from_pretrained(
78
+ repo, subfolder="metallic_transmission_mapper", strict=True
79
  ).eval()
80
  ```
81