pxGeniusAI/lama-dilated

LAMA Dilated Inpainting Model (ONNX)

LAMA (Large Mask Inpainting) model with dilated convolutions, exported to ONNX format for browser-based
inpainting via ONNX Runtime Web (WebGPU backend).

Model Details

• Architecture: LAMA with dilated convolutions
• Format: ONNX (FP32)
• Size: ~174MB
• Input: 512×512 RGB image + 512×512 binary mask
• Output: 512×512 inpainted RGB image
• Runtime: ONNX Runtime Web with WebGPU or WASM backend

Usage

This model is loaded directly in the browser for local-mode text removal and inpainting on images.

Direct download URL

https://huggingface.co/pxGeniusAI/lama-dilated/resolve/main/lama_dilated.onnx

Source

Exported from Qualcomm AI Hub — LaMa-Dilated, which is based on the original LAMA model by Samsung AI Center.

License

BSD-3-Clause