Question/Issue:
I have utilized 90 images and labeled them using -Bounding box labeling with OWL-ViT
I was able to generate features successfully
When I try to train the images using FOMO I am facing errors in the console
Project ID:
561596
Context/Use case:
I am facing issue at object detection task while training the images
Expected Outcome:
Expected to successfully train my images
Hi @brock - the team did identify a bug and will be working to resolve it.
As an aside, I took a look at your project and wanted to let you know that FOMO is not the best architecture for the data you are using. FOMO is a really cool architecture as it brings object detection to highly resource constrained devices such as MCUs. On the flip side, however, to make this happen there were some novel architecture decisions, mainly around the output grid and using object centroids, that led to some limitations.
FOMO works best when:
There are no overlapping object centroids
The size of the bounding box for an object is roughly the same size as an output grid cell
This is not the case for your data. You can read more about FOMO in the docs here and in the blog here.
