Web version support for labeling

[AmartyaCSB]

Thanks for the fantastic research and the code. It is very beneficial as it supports zero shot segmentation

Can we use the pre-trained models for custom images? Researchers can also use this for labeling their images if we can use the web tool and perform labeling and create a JSON/COCO format file. Any ideas?

[zhongwei]

import cv2, json
from segment_anything import SamAutomaticMaskGenerator, sam_model_registry

sam = sam_model_registry['vit_b'](checkpoint='sam_vit_b_01ec64.pth')
generator = SamAutomaticMaskGenerator(sam, output_mode='coco_rle')

image = cv2.imread('punch_out.jpg')
image = cv2.cvtColor(image, cv2.COLOR_BGR2RGB)

masks = generator.generate(image)
with open('sample_segment.json', "w") as f:
    json.dump(masks, f)

[lujiazho]

Here is a simple implementation of web-based mask-drawer if interested: https://github.com/lujiazho/SegDrawer

[Kingfish404]

Yet another SAM web UI https://github.com/kingfish404/segment-anything-webui

[DuinoDu]

CVAT also supports SAM, FYI.