search

hustvl / EVF-SAM

Official code of "EVF-SAM: Early Vision-Language Fusion for Text-Prompted Segment Anything Model"
Apache License 2.0
311 stars 13 forks source link

objects365 dataset annotation file #27

Open GinnyXiao opened 1 month ago

GinnyXiao commented 1 month ago

Dear authors,

Thanks for releasing the updated checkpoints for multi-task segmentation! I tried out the semantic level segmentation demo, and the results are greatly improved.

I was wondering if you'd share the objects365 dataset version you are using to train the checkpoints? And where can I find the annotation files (train & val) that contain segmentation masks? Much appreciated!

CoderZhangYx commented 1 month ago

We select from original o365 annotations by excluding categories with more than one instance for each image. Then we apply sam-2 to transfer bounding boxes to segmentation masks. Easy code honestly. It might take few hours to annotate the full dataset. We will present detail in our next update of paper, and will release the annotation code.