Support YOLOv8 Ins Segmentation Inference

[wkpark]

Don't know why but PR #636 already merged onto a separate dev branch but was accidentally removed without reason, I guess.

rebased/squashed and conflict resolved manually onto the main branch #636

original works done by @Nioolek

---

• works nicely with converted yolov8 segment models. (2023/09/28)

Motivation

Please see #636

Modification

Support YOLOv8 Ins Segmentation Inference

Checklist

1. Pre-commit or other linting tools are used to fix potential lint issues.
2. The modification is covered by complete unit tests. If not, please add more unit tests to ensure the correctness.
3. If the modification has a potential influence on downstream projects, this PR should be tested with downstream projects, like MMDetection or MMClassification.
4. The documentation has been modified accordingly, like docstring or example tutorials.

[CLAassistant]

All committers have signed the CLA.

[lijoe123]

@wkpark 请问，为什么我在使用你仓库里的yolov8_ins时，会出现 报错

File "/home/music/Downloads/mmyolo-sam/mmyolo/models/dense_heads/yolov5_head.py", line 470, in loss
    losses = self.loss_by_feat(*loss_inputs)
TypeError: loss_by_feat() takes from 6 to 7 positional arguments but 8 were given

[1dmesh]

It's odd, as it looks like the changes for transforms.py from the original PR stayed in main as well in some form or fashion.