Skip to content

segmentation with yolov5 #508

@rhsaeedy

Description

@rhsaeedy

I'm following this tutorial to design a custom instance segmentation on OAK-D. I trained the custom model with only one class and used depthai online tool to convert it to the blob. When I use main.py from host-decoding folder I get the following error:

Traceback (most recent call last):
  File "~/depthai-experiments/gen2-yolo/host-decoding/main.py", line 170, in <module>
    boxes = non_max_suppression(output, conf_thres=conf_thresh, iou_thres=iou_thresh)
  File "~/depthai-experiments/gen2-yolo/host-decoding/util/functions.py", line 29, in non_max_suppression
    xc = prediction[..., 4] > conf_thres  # candidates
IndexError: index 0 is out of bounds for dimension 1 with size 0

I also tried other values instead of 4 in xc = prediction[..., 4] and got the same error.
I trained the NN with 640x640 image size. For now I really want to pass this error but my final goal is to be able to get the polygon masks that I used to label the data with for instance segmentation. I wonder if it is possible or not?
By the way, when I print the prediction.shape I get torch.Size([1, 10647, 0])

Metadata

Metadata

Assignees

No one assigned

    Labels

    No labels
    No labels

    Type

    No type

    Projects

    No projects

    Milestone

    No milestone

    Relationships

    None yet

    Development

    No branches or pull requests

    Issue actions