AXERA-TECH/EdgeTAM

By AXERA-TECH · ↓ 5 · ♥ 0 · Updated 2026-01-30

What's in this repo

This page lists the files in AXERA-TECH/EdgeTAM, including edgetam_image_encoder.onnx, edgetam_mask_decoder.onnx, edgetam_prompt_encoder.onnx and more.

4 ONNX models (.onnx files)
29 files, including 5 .py, 5 .png, 4 .axmodel, 4 .onnx, 3 .jpg, 2 .npy
Last updated 2026-01-30

ONNX files

File	Size	Download
onnx_models/edgetam_image_encoder.onnx	—	Download
onnx_models/edgetam_mask_decoder.onnx	—	Download
onnx_models/edgetam_prompt_encoder.onnx	—	Download
onnx_models/edgetam_prompt_mask_encoder.onnx	—	Download

How to use

These ONNX files can be loaded with ONNX Runtime. To inspect a model graph, upload the file to the Netron viewer. Download links point to the original files on Hugging Face.

README

EdgeTAM

基于EdgeTAM的图像分割Pipeline，支持多种输入提示（框、点、掩码），支持650N系列平台的模型推理。

支持芯片:

AX650N

支持硬件

原始模型请参考

性能对比

输入图片大小 1024x1024

Models	Latency (ms)	CMM Usage (MB)
edgetam_image_encoder	22.348	29.124
edgetam_prompt_encoder	0.055	0.023
edgetam_prompt_mask_encoder	0.457	0.037
edgetam_mask_decoder	4.729	16.730

模型转换

模型转换工具链Pulsar2
转换文档Model Convert

环境准备

NPU Python API: pyaxengine

安装需要的python库 pip install -r requirements.txt

运行

(myenv) root@ax650:~/EdgeTAM# python3 image_prediction_ax.py --input_box 75,275,1725,850
[INFO] Available providers:  ['AxEngineExecutionProvider']
Loading EdgeTAM Onnx models...
[INFO] Using provider: AxEngineExecutionProvider
[INFO] Chip type: ChipType.MC50
[INFO] VNPU type: VNPUType.DISABLED
[INFO] Engine version: 2.12.0s
[INFO] Model type: 2 (triple core)
[INFO] Compiler version: 5.0-patch1-dirty a512c95e-dirty
[INFO] Using provider: AxEngineExecutionProvider
[INFO] Model type: 2 (triple core)
[INFO] Compiler version: 5.0-patch1-dirty a512c95e-dirty
[INFO] Using provider: AxEngineExecutionProvider
[INFO] Model type: 2 (triple core)
[INFO] Compiler version: 5.0-patch1-dirty a512c95e-dirty
[INFO] Using provider: AxEngineExecutionProvider
[INFO] Model type: 2 (triple core)
[INFO] Compiler version: 5.0-patch1-dirty a512c95e-dirty
Get prompts:
  input_box: [  75  275 1725  850]
  input_point_coords: None
  input_point_labels: None
Only box input provided
Get dense_embeddings_no_mask
[0.9777304]
✅ Saved: ./results/mask_1.png

保存结果在 ./results 目录下：

(myenv) root@ax650:~/EdgeTAM# python3 image_prediction_ax.py --image_path ./examples/images/truck.jpg --input_box 425,600,700,875 --input_point_coords 575,750 --input_point_labels 0
[INFO] Available providers:  ['AxEngineExecutionProvider']
Loading EdgeTAM Onnx models...
[INFO] Using provider: AxEngineExecutionProvider
[INFO] Chip type: ChipType.MC50
[INFO] VNPU type: VNPUType.DISABLED
[INFO] Engine version: 2.12.0s
[INFO] Model type: 2 (triple core)
[INFO] Compiler version: 5.0-patch1-dirty a512c95e-dirty
[INFO] Using provider: AxEngineExecutionProvider
[INFO] Model type: 2 (triple core)
[INFO] Compiler version: 5.0-patch1-dirty a512c95e-dirty
[INFO] Using provider: AxEngineExecutionProvider
[INFO] Model type: 2 (triple core)
[INFO] Compiler version: 5.0-patch1-dirty a512c95e-dirty
[INFO] Using provider: AxEngineExecutionProvider
[INFO] Model type: 2 (triple core)
[INFO] Compiler version: 5.0-patch1-dirty a512c95e-dirty
['575,750']
575,750
Get prompts:
  input_box: [425 600 700 875]
  input_point_coords: [[575 750]]
  input_point_labels: [0]
Get dense_embeddings_no_mask
[0.90291053]
✅ Saved: ./results/mask_1.png

Model info

image-segmentation

onnx ONNX image-segmentation base_model:facebook/EdgeTAM base_model:quantized:facebook/EdgeTAM license:afl-3.0 region:us

↓ 5 ♥ 0

Created: 2025-12-01

Updated: 2026-01-30

View on Hugging Face

Files (29)

.gitattributes

README.md

axmodel/dense_embeddings_no_mask.npy

axmodel/edgetam_image_encoder.axmodel

axmodel/edgetam_mask_decoder.axmodel

axmodel/edgetam_prompt_encoder.axmodel

axmodel/edgetam_prompt_mask_encoder.axmodel

config.json

examples/images/cars.jpg

examples/images/groceries.jpg

examples/images/truck.jpg

image_prediction_ax.py

image_prediction_onnx.py

onnx_models/dense_embeddings_no_mask.npy

onnx_models/edgetam_image_encoder.onnx ONNX

onnx_models/edgetam_mask_decoder.onnx ONNX

onnx_models/edgetam_prompt_encoder.onnx ONNX

onnx_models/edgetam_prompt_mask_encoder.onnx ONNX

requirements.txt

results/mask_1.png

results/mask_2.png

results/mask_3.png

results/mask_4.png

results/mask_5.png

utils/EdgeTAM_image_predictor.py

utils/EdgeTAM_image_predictor_onnx.py

utils/__pycache__/EdgeTAM_image_predictor.cpython-311.pyc

utils/__pycache__/transforms.cpython-311.pyc

utils/transforms.py

Related models

More image-segmentation models

Browse all image-segmentation models