sheldon-97 commited on
Commit
d77be8c
·
verified ·
1 Parent(s): cd5c97e

Update README.md

Browse files
Files changed (1) hide show
  1. README.md +20 -24
README.md CHANGED
@@ -1,5 +1,5 @@
1
  ---
2
- license: apache-2.0
3
  library_name: diffusers
4
  pipeline_tag: image-to-image
5
  tags:
@@ -8,33 +8,29 @@ tags:
8
  - viton-hd
9
  ---
10
 
11
- # DualVTOFF - Official Model of the paper "What Matters in Virtual Try-Off? Dual-UNet Diffusion Model For Garment Reconstruction" (ICPR 26)
12
 
13
- DualVTOFF reconstructs a canonical frontal garment image from a person image and
14
- a binary mask of the garment region. This release contains separate 512 x 384
15
- and 1024 x 768 checkpoints. It is intended for research and development.
16
 
17
- The checkpoints were trained for the VITON-HD domain. Performance outside that
18
- domain, across garment categories, body types, image conditions, and demographic
19
- groups has not been systematically characterized in this release. Generated
20
- images can contain incorrect texture, shape, text, logos, or color. Users should
21
- validate outputs for their application and should not use them as evidence of
22
- real products or people.
23
 
24
- Recommended inference is fp16 CUDA, 30 denoising steps, and guidance scale 2.0.
25
- Inputs are resized directly to the chosen native resolution. The mask uses white
26
- for the source garment and black elsewhere. See `README.md` for commands.
27
-
28
- ## Inference code:
29
-
30
- Refer to : [https://github.com/locphattruong/dual-vtoff](https://github.com/locphattruong/dual-vtoff)
31
 
32
  ## Licenses
33
 
34
- Repository-authored code is distributed under `LICENSE`. Vendored Diffusers and
35
- IP-Adapter-derived files retain their upstream Apache-2.0 headers/terms. SDXL
36
- base components are subject to CreativeML Open RAIL++-M. This repository does
37
- not assert that the MIT code license grants rights to the checkpoint weights;
38
- the publisher must attach the final applicable weight license before upload.
39
-
40
 
 
 
 
 
 
1
  ---
2
+ license: openrail++
3
  library_name: diffusers
4
  pipeline_tag: image-to-image
5
  tags:
 
8
  - viton-hd
9
  ---
10
 
11
+ # DualVTOFF checkpoints
12
 
13
+ Official checkpoints for *What Matters in Virtual Try-Off? Dual-UNet Diffusion
14
+ Models for Garment Reconstruction*. DualVTOFF reconstructs a canonical garment
15
+ image from a person image and a binary garment-region mask.
16
 
17
+ | Checkpoint | Output height x width |
18
+ | --- | --- |
19
+ | [VITON-HD-512-res](VITON-HD-512-res/) | 512 x 384 |
20
+ | [VITON-HD-1024-res](VITON-HD-1024-res/) | 1024 x 768 |
 
 
21
 
22
+ See the [inference code and instructions](https://github.com/locphattruong/dual-vtoff).
23
+ Each checkpoint includes its inference components, manifest, checksums, license,
24
+ and attribution notices.
 
 
 
 
25
 
26
  ## Licenses
27
 
28
+ The checkpoint weights in this directory are released under **CreativeML Open
29
+ RAIL++-M**, including the use restrictions in Attachment A. The full license is
30
+ included in [LICENSE](LICENSE). Users and downstream distributors must comply
31
+ with its use, attribution, and redistribution conditions.
 
 
32
 
33
+ The inference source code is released separately under Apache-2.0. Component
34
+ copyright and license notices are preserved in [NOTICE.md](NOTICE.md) and
35
+ [licenses/](licenses/). The code license does not replace this model license.
36
+ VITON-HD data is not included and remains subject to its separate dataset terms.