Aller au contenu
ModelPrésentéExterne
Browse files

Triposr Onnx Web

Single-image to 3D reconstruction. TripoSR exported to ONNX (fp16) and split into three sessions so it runs entirely in the browser on onnxruntime-web with the WebGPU execution provider (GridSample runs natively).

Sourcehuggingface.co/dcharlot65-aurasense/triposr-onnx-web/tree/e23007c3ce90bb968eae014c0168981871fe2d0d
TypeOtherVersionBackbone fp16CClasse

Source du modèle

Type
Other
Taille du fichier
95,4 KB–636 MB
Version
vBackbone fp16
Mis à jour
10 oct. 2026
Source
Hugging Face

Description de la source

TripoSR fp16 ONNX for browser WebGPU

Single-image to 3D reconstruction. TripoSR exported to ONNX (fp16) and split into three sessions so it runs entirely in the browser on onnxruntime-web with the WebGPU execution provider (GridSample runs natively).

Files:

  • encoder_fp16.onnx (~173 MB) - ViT image encoder. Input image (1,3,512,512) fp32 -> tokens (1,768,1025).
  • backbone_fp16.onnx (~666 MB) - triplane transformer. Input image_tokens (1,1025,768) -> scene_codes (1,3,40,64,64).
  • decoder_fp16.onnx (~0.1 MB) - triplane MLP. Inputs triplane (3,40,64,64) + points (N,3) -> density (N,1), color (N,3).

Model I/O is fp32; weights are fp16. Preprocess: RGBA composited over 0.5 gray, bilinear antialias resize to 512x512. Decode a grid over (-0.87, +0.87), marching cubes at iso 25 on the density field.

Credit

Original model: VAST-AI-Research/TripoSR (Tripo AI + Stability AI). Apache-2.0. This is a repackaging of those weights to ONNX for in-browser inference.

Sources

1 source
Hugging FacePrincipalVersion e23007c3ce90bb968eae014c0168981871fe2d0d
huggingface.co/dcharlot65-aurasense/triposr-onnx-web/tree/e23007c3ce90bb968eae014c0168981871fe2d0d

Vérifié 10 oct.

Download files

Artefacts du modèle

3 artefacts

backbone_fp16.onnx

onnx · 636 MB · SHA-256 d0bc00a5f6a7…edc7 · Hugging Face

Download file

decoder_fp16.onnx

onnx · 95,4 KB · SHA-256 20315f1ddb74…7d99 · Hugging Face

Download file

encoder_fp16.onnx

onnx · 165 MB · SHA-256 494063a59595…c966 · Hugging Face

Download file

Extraits de sources

2 extraits
Readme1 aoûthuggingface.co/dcharlot65-aurasense/triposr-onnx-web

--- license: apache-2.0 tags: - triposr - image-to-3d - onnx - webgpu - onnxruntime-web library_name: onnxruntime --- # TripoSR fp16 ONNX for browser WebGPU Single-image to 3D reconstruction. TripoSR exported to ONNX (fp16) and split into three sessions so it runs entirely in the browser on onnxruntime-web with the WebGPU execution provider (GridSample runs natively). Files: - `encoder_fp16.onnx` (~173 MB) - ViT image encoder. Input `image` (1,3,512,512) fp32 -> `tokens` (1,768,1025). - `backbone_fp16.onnx` (~666 MB) - triplane transformer. Input `image_tokens` (1,1025,768) -> `scene_codes` (1,3,40,64,64). - `decoder_fp16.onnx` (~0.1 MB) - triplane MLP. Inputs `triplane` (3,40,64,64) + `points` (N,3) -> `density` (N,1), `color` (N,3). Model I/O is fp32; weights are fp16. Preprocess: RGBA composited over 0.5 gray, bilinear antialias resize to 512x512. Decode a grid over (-0.87, +0.87), marching cubes at iso 25 on the density field. ## Credit Original model: [VAST-AI-Research/TripoSR](https://github.com/VAST-AI-Research/TripoSR) (Tripo AI + Stability AI). Apache-2.0. This is a repackaging of those weights to ONNX for in-browser inference.

Note1 aoûthuggingface.co/dcharlot65-aurasense/triposr-onnx-web

Source context: 0 downloads · 0 likes · Pipeline image-to-3d · Library onnxruntime · Repo dcharlot65-aurasense/triposr-onnx-web