Open renyiyu opened 1 year ago
This issue was also mentioned, and I'm guessing it's a label studio issue, the exact problem needs further troubleshooting.
I solved this issue by loading the model only one time if it has not been loaded before rather than loading the model by labeling each image.
Open /playground/label_anything/sam
and open the mmdetection.py and go to the class MMDetection
and change it as follows:
class MMDetection(LabelStudioMLBase):
"""Object detector based on https://github.com/open-mmlab/mmdetection."""
_predictor = None # ----------> add this private variable
def __init__(self,
config_file=None,
checkpoint_file=None,
sam_config='vit_b',
sam_checkpoint_file=None,
image_dir=None,
labels_file=None,
out_mask=True,
out_bbox=False,
out_poly=False,
score_threshold=0.5,
device='cpu',
**kwargs):
super(MMDetection, self).__init__(**kwargs)
#************************Add the following two lines*******************
# Only load the model if it hasn't been loaded before
if MMDetection._predictor is None:
MMDetection._predictor = load_my_model(device, sam_config, sam_checkpoint_file)
self.PREDICTOR = MMDetection._predictor
Label Studio X SAM seems to have memory issues after auto annotating for about 20-30 objects in the same image. The auto annotating process is getting slower and slower even I use vit_b/mobile sam model on gpu setup.
My average image size is around 640 * 640 and my GPU is 3090. Since my use case has around 100 small objects to annotate per image. It makes me really difficult to annotate all the objects for the entire image, especially last 50 objects.