search

SysCV / sam-hq

Segment Anything in High Quality [NeurIPS 2023]
https://arxiv.org/abs/2306.01567
Apache License 2.0
3.7k stars 223 forks source link

The val_iou and val_boundary_iou is very low #107

Open Ghy1209 opened 11 months ago

Ghy1209 commented 11 months ago

Validating... valid_dataloader len: 131 [ 0/131] eta: 0:01:00 val_iou_0: 0.4816 (0.4816) val_boundary_iou_0: 0.4124 (0.4124) time: 0.4584 data: 0.1234 max mem: 18932 [130/131] eta: 0:00:00 val_iou_0: 0.0486 (0.1484) val_boundary_iou_0: 0.0486 (0.1415) time: 0.3432 data: 0.0013 max mem: 18932 Total time: 0:00:44 (0.3411 s / it)

How can I solve it?

xuans29 commented 2 months ago

on the opposite: RuntimeWarning: invalid value encountered in scalar divide boundary_iou = intersection / union /home/dell/desktop/waterbody/sam_hq/sam-hq-main/train/utils/misc.py:536: RuntimeWarning: invalid value encountered in scalar divide boundary_iou = intersection / union [383/384] eta: 0:00:00 val_boundary_iou_0: nan (nan) val_iou_0: nan (nan) time: 0.4104 data: 0.0030 max mem: 18919 Total time: 0:02:38 (0.4117 s / it)

Averaged stats: val_boundary_iou_0: nan (nan) val_iou_0: nan (nan)

i dont know if my train data matters bracause i am trying to finetune SAM to a Binary Classification Task