I confirmed that each image and labeling worked normally in the Annotate step.
However, after setting options 3 and 4 and 5 in the generation stage, the results checked by versions are always incorrectly labeled. What is the reason that I didn’t choose any option because I thought it was because I made the wrong choice in option 3 and option 4 and 5 of generator, but the result is the wrong data??
To clarify, the problem you’re facing is that the bounding boxes in the generated version are displaced correct? Can you try generating a new version and seeing if the problem persists?
If that’s the case, could you add Roboflow Support to your workspace so I can troubleshoot this? You can toggle Roboflow Support on the invite page.
@Emily
So I think their problem is that:
In the “Annotate” step, they confirmed that both the images and their corresponding labels were functioning correctly. However, in the subsequent “Generation Stage,” after configuring options 3, 4, and 5, the results they are checking always have incorrect labels. Although I don’t know what options 3, 4 and 5 are relating to exactly.
I see that in the two pictures they sent, they show that the first set of images is properly labeled, it is the training data they used for the model, properly annotated. In the second image they showed we can see that it appears to be model predictions, and they appear to be off. I think maybe @Yangho_KIM_김양호_DTWORESOURCE thought that in the second image they would all be perfectly labeled like the training data in the first image, thus the confusion.
Even if I try to create a version by changing the options at each stage of generate, the wrong bounding box always results in the wrong bounding box.I don’t think there will be any wrong cases because that stage is not a learning stage, but I think it’s weird that the box keeps drawing outside the range I choose.
I added robotflow support. I hope the data can be generated properly. Thank you.
I was able to check out your workspace and it looks like the auto-orient setting was turned off during preprocessing. I generated a new version with the auto-orient on and that fixed the issue.
You can learn more about what auto-orient does on our blog post, happy building!