Name		Name	Last commit message	Last commit date
assets		assets
configs/cls/imagenet		configs/cls/imagenet
efficientvit		efficientvit
sample		sample
.gitignore		.gitignore
EfficientViT.md		EfficientViT.md
LICENSE		LICENSE
README.md		README.md
TRAINING.md		TRAINING.md
convert.py		convert.py
demo_sam_model.py		demo_sam_model.py
eval_cls_model.py		eval_cls_model.py
eval_sam_coco.py		eval_sam_coco.py
eval_seg_model.py		eval_seg_model.py
logger.py		logger.py
onnx_export.py		onnx_export.py
pyproject.toml		pyproject.toml
requirements.txt		requirements.txt
sam_export.py		sam_export.py
sam_onnx_infer.py		sam_onnx_infer.py
tflite_export.py		tflite_export.py
train_cls_model.py		train_cls_model.py

Repository files navigation

End-to-end SAM ONNX export for EfficientViT

Include input preprocess(resizing, NHWC to NCHW) and output mask postprocess in a single model.

The input image,and point prompt input should include batch axis.

The output mask has same shape as an original input image.

Export usage

python sam_export.py --model {model_type} --checkpoint {path_to_model_weight} --output {output_path} --include-resize

If you don't want to incldue pre/postprocessing step in the ONNX model, just omit --include-resize.

TODO

code refactoring
implement auto mask generation

About

EfficientViT is a new family of vision models for efficient high-resolution vision.

1 star

0 watching

Report repository

Releases

Packages

Contributors

Languages