Lower Resolution Model?

[3DTOPO]

I got DepthPro model inferencing on my iPad Pro but uses up 5GB of RAM which is just not feasible on many devices.

I've tried adding flexible shapes and creating a half res model without any luck (weight mismatch). Any chance we could see a lower resolution model?

[wzjoriv]

I would recommend hosting it on a server and allowing for API calls from less powerful devices. You could also try distillation techniques

[3DTOPO]

Hosting the model is not possible for the business model.

[Feiyu-Liu]

Have you already converted the model to the coreml format? I used a converted model from huggingface, but I encountered severe memory leaks in the simulator, and I haven't pinpointed the issue yet.😢