BEN2 — Background Erase Network 2 (ONNX, FP32)
FP32 conversion of the BEN2 FP16 ONNX model for CPU and older GPU compatibility.
The official onnx-community repo only provides an FP16 variant, which requires a modern GPU. This FP32 version runs on CPU and older GPUs (e.g. NVIDIA Maxwell/Kepler, Intel HD).
Model Specs
| Property | Value |
|---|---|
| File size | ~403 MB |
| Input shape | (1, 3, 1024, 1024) float32, ImageNet-normalized |
| Output shape | (1, 1, 1024, 1024) float32 |
| Postprocessing | Min-max normalization (not sigmoid) |
| Original | onnx-community/BEN2-ONNX |
Preprocessing
Same as BEN2 FP16: resize to 1024x1024, scale to [0,1], apply ImageNet normalization (mean=[0.485, 0.456, 0.406], std=[0.229, 0.224, 0.225]).
Conversion
Converted from the FP16 model using ONNX Python API — all FP16 initializers, inputs, outputs, and intermediate value_info cast to FP32.
License
Apache-2.0 — same as the original BEN2 model.
Acknowledgments
- Original model by PramaLLC/BEN2
- FP16 ONNX export by onnx-community
- FP32 conversion by ListingGems — an offline product photo editor for marketplace sellers
Model tree for Pezhgorski/BEN2-FP32-ONNX
Base model
PramaLLC/BEN2