I am working on Fine Tuning a SAM2 model as instructions given by the blog How to Fine-Tune SAM-2.1 on a Custom Dataset
The problem is I did not have the annotated data. I had the SVG paths. So, i have used a python code to convert those SVG images into images.jpg and images.json as suggested by the blog. I have used the RLE segmentation to convert those polygons of SVG path into pixel level masks. I have used 797 images for fine tunning.
The process is done for Home Exterior segmentations like Piller, Base, Projection etc. The results i get is not useful as i not get anything from the result.
I am attaching the result here:
Kindly tell me why this happens?