Are there recommended input shape for each model?

[tucan9389]

Hi, I'm an iOS app developer who is interesting in ML.
Firstly, thank you for sharing your awesome vision modeling project.

I want to support imgclsmob's pre-trained models on my repositories which provide baseline code for inferencing various ML models on iOS devices.

I converted all imgclsmob's pre-trained pose estimation model to tflite with (1,224,224,3) input shape. But I think each model has an optimized input shape that was used on training.

Are there recommended input shape for each model?

Here is the issue what I'm working:
tucan9389/PoseEstimation-TFLiteSwift#59

Thank you.

[osmr]

Nice work!
To answer your question, see the source code of each model, in particular, the default value of the in_size argument. In most cases it's (1, 256, 192, 3).

[tucan9389]

• 256: height
• 192: width

Right?

Now I’m testing all pose models provided by this repo. When I’m done, I’ll share it.

Thank you!