Questions about Roboflow Trained model! and i want to know why my own trained model is suck

Hi I have some questions about roboflow trained model.
I created a model from roboflow website and the model was very good.
However, I encountered some problems.

  1. I could not download the model to my local so I tried to train by myself through Ultrayics. However, even though I used same image sets, the result was absolutely different. My own model also had pretty good mAP but its real performance was very poor but roboflow model performance was very good. Do you know why there is a difference? and may i know your guys magics on training?

  2. This is real problem. since my model was uselessly bad, I tried to use roboflow trained model api to predict my video from the local. However, since it was api call, it was very slow. It is taking over 1sec to predict 1frame… Is there any way I can improve prediction??

Please help me!!!
I want to make my model better like roboflow trained model or I want to get fast response from the model api from roboflow!!

Hey @Hs_L

Sorry to hear you’re having trouble training with Ultralytics, but we don’t provide support for that here.

Unfortunately, we do not share the architecture or hyperparameters behind Roboflow Train.

Let’s try to get that solved for you. I was able to reproduce your issue and we’re looking into it, but in the meantime, you can try using roboflow.js (powering the webcam demo which seems to be performing fine) or run your model locally using our inference package.

1 Like

Hi @Hs_L

Looking into this issue, I was able to replicate this issue on the Universe model page, but not on the Hosted API or anywhere else. Has this been your experience?

1 Like

Actually I used my other roboflow id to train and predict the model.

This topic was automatically closed 14 days after the last reply. New replies are no longer allowed.