<div class="snippet-clipboard-content notranslate position-relative overflow-auto" data-snippet-clip

Can I use it to evaluate other segmentation datasets which is not included in the mmsegmentation about beyond-iou HOT 2 CLOSED

yanhaijun11902 commented on June 27, 2024

Can I use it to evaluate other segmentation datasets which is not included in the mmsegmentation

from beyond-iou.

Comments (2)

yanhaijun11902 commented on June 27, 2024 1

Got it ,I am deeply indebted to you for your help.

from beyond-iou.

mxbh commented on June 27, 2024

Hi,
thanks for your interest.

Here's a snippet showing how you can use our method on your binary dataset:

from beyond_iou import Evaluator, build_segmentation_loader, Result

evaluator = Evaluator(
    class_names=['background', 'foreground'],
    ignore_index=255, # optionally modify for your case if there are ignore pixels
    boundary_width=0.01, # choose reasonably for your case
)
loader = build_segmentation_loader(
    pred_dir='path/to/your/predictions', # as 0-1 images, otherwise use pred_label_map to map label ids accordingly
    gt_dir='path/to/your/groundtruth', # as 0-1 images, otherwise use gt_label_map to map label ids accordingly
    gt_label_map=None,
    pred_label_map=None,
    num_workers=0,
    pred_suffix='',
    gt_suffix=''
)
result = evaluator.evaluate(loader)
result.create_report('your/output/directory', exist_ok=True)

This treats background (0) and foreground (1) as separate classes and computes the metrics as the mean over the two classes. If you are interested in the results for the foreground class only, you can access them through result.dataframe.

from beyond-iou.

Recommend Projects