ANNOUNCEMENT: Custom Model Weights Upload Overview

There have been some reports of issues, to clear them up:

  1. If your model weights uploads fails, check to see if you’ve set the imgsz parameter in the training code.
  • Example: resized images to 640x640 prior to export? Set imgsz=640.
  • If you didn’t resize, check your median image size in the Dataset Health Check tool. Resize your images to a square size based on this value, and set that value for imgsz
  • If you didn’t resize, but unsure what to set your resize value to, I recommend generating a version with 640x640 resize, as the model will train faster, and process predictions faster, with the smaller input size.
  1. Model weights uploads, at this time of this posting, are only available on Roboflow for YOLOv5 Object Detection, YOLOv8 Object Detection, and YOLOv7 Instance Segmentation.
  • We will make announcements as more models are added.

Please continue new reports of issues on this thread.

Hi Mohamed, now the deployment worked but I still face issues:
The result of the training (mAP, Precision, Recall) does not appear in the Roboflow Deployment is way worse within Roboflow. So somehow within the deployment the results get lost/are replaced?

Result in Collab:

Result of same model deployed in roboflow:

Hi @Ari_Q, do you have a copy of the notebook you used in training?

Can you send an email to with:

  1. A link to your shared Colab notebook, if you still have it saved in your Google Drive
  2. The link to the affected dataset version
  3. The email for your account

Hi @Mohamed, I just sent you a mail with the wished infos. Looking forward to your feedback.

Thank you - I’ve received it and I’m looking now

There are similar issues

Training results

Result of same model deployed in roboflow:

Label Assist cannot be used for labeling