Open Ahmed14974 opened 1 year ago
Would appreciate any information here. Thanks!
Hi @Czm369, would you have any info about the query above? Please let me know. Thanks!
@Czm369 just wanted to follow-up on this query. Please provide the relevant information. Thanks!
For Mask-RCNN and Solov2 Instance Segmentation models, what is the correct way to handle negative images (images where the objects desired to be detected are not present)? For the negative images, in the COCO format json file, I have an entry in the image dict, but nothing in the annotation dict. For such images (where the annotation dict does not have an element), are they completed excluded from training in Mask-RCNN and Solov2 current implementation?