search

ibaiGorordo / ONNX-SAM2-Segment-Anything

Python scripts for the Segment Anythin 2 (SAM2) model in ONNX
MIT License
165 stars 9 forks source link

How to get all masks directly? #14

Open MonolithFoundation opened 1 month ago

MonolithFoundation commented 1 month ago

How to get all masks directly?

Matagi1996 commented 1 month ago

The automatic mask predictor is sampling a grid of points and calling the decoder again and again, I have actually tried this with the Onnx model (not implemented here) but found it quite slow compared to pytorch, I guess because the decoder does not use flash attn 2 when exported to onnx and this adds up if you inference a few hundret times / image to get all masks.

MonolithFoundation commented 1 month ago

Have u referenced another sam2 onnx implementation? looks like they make all works, including video tracking.