I am attempting to train a Faster R-CNN model using the HAM10000 dataset. However, I have been unable to locate an annotation file specifically for this dataset. I am seeking guidance on the most effective approach to annotate bounding boxes for this dataset. Is there any possibility of automating this process to streamline the annotation procedure?
$\begingroup$
$\endgroup$
Add a comment
|
1 Answer
$\begingroup$
$\endgroup$
2
Mysuggestion would be this: manually annotate few hundreds samples and then train a yolo5 model or SSD model to detect and classify objects
then from the model you can create the annotation for the rest of images then check the annotation that made by the model and fix the errors
finally fint-une the model on the final dataset
-
$\begingroup$ Thank you! I employed a pre-trained U-Net segmentation model to segment the data and extract bounding boxes from the segmentation. The obtained results were promising. $\endgroup$ Commented Jul 6, 2023 at 8:30
-
$\begingroup$ Good to hear, see also the tool segment anything $\endgroup$ Commented Jul 6, 2023 at 8:39