Does the exported SAM2 encoder and decoder only support inference images？

[Aimol-l]

I plan to conduct video inference, but it seems that having only encoder and decoder is not enough.
Perhaps it includes：Image encoder，Memory attention. Prompt encoder and mask decoder，Memory encoder，Memory bank.

I'm a bit confused.😕

[vietanhdev]

Hi @Aimol-l ,
There is no inference code at the moment. Please use the official repo from Meta.
Thank you!

[MELANCHOLY828]

我打算做视频推理，但是好像只有encoder和decoder还不够， 可能还要包括：Imagecoder，Memory Attention，Prompt Encoder和Mask Decoder，Memory Encoder，Memory Bank。

我有点困惑。😕

I also want to use it for video inference. Does the exported SAM2 encoder and decoder support it?

[Aimol-l]

我打算做视频推理，但是好像只有encoder和decoder还不够， 可能还要包括：Imagecoder，Memory Attention，Prompt Encoder和Mask Decoder，Memory Encoder，Memory Bank。

我有点困惑。😕

I also want to use it for video inference. Does the exported SAM2 encoder and decoder support it?

I am try to export Memory_attention module,but failed. this is my code ：https://github.com/Aimol-l/SAM2Export.git
so,I'm stuck。
this is my note graph.

[MELANCHOLY828]

我打算做视频推理，但是好像只有encoder和decoder还不够， 可能还要包括：Imagecoder，Memory Attention，Prompt Encoder和Mask Decoder，Memory Encoder，Memory Bank。

我有点困惑。😕

I also want to use it for video inference. Does the exported SAM2 encoder and decoder support it?

I am try to export Memory_attention module,but failed. this is my code ：https://github.com/Aimol-l/SAM2Export.git so,I'm stuck。 this is my note graph.

你好，请问图片预测的运行成功了吗，我报错
Traceback (most recent call last):
File "", line 5, in
File "e:\anaconda\envs\sam2\lib\site-packages\imread_from_url\imread_from_url.py", line 52, in imread_from_url
image = cv2.cvtColor(np.array(image, dtype=np.uint8), cv2.COLOR_RGB2BGR)
cv2.error: OpenCV(4.10.0) 👎 error: (-5:Bad argument) in function 'cvtColor'

Overload resolution failed:

• src is not a numpy array, neither a scalar
• Expected Ptrcv::UMat for argument 'src'
  不知道是不是版本问题

[vietanhdev]

Closing this as answered and documented in v0.5.0: these split SAM2 encoder/decoder graphs support image prediction only. Video inference requires the temporal memory attention, memory encoder, and memory bank from the official tracker, which are intentionally outside the current export contract.