当前位置：首页 > news >正文

大模型SAM辅助labelme分割数据集（纯小白教程）

news 来源：原创 2025/5/2 16:01:13

一、labelme的安装

1.打开终端（win+r，输入cmd）,输入以下命令

2.创建虚拟环境

conda create -n data python==3.10.0

3.激活虚拟环境

conda activate data

4.安装labelme

pip install labelme==5.6.1

出现下面内容则安装成功

5.AI辅助标注

从以下路径中下载不同模型的ONNX文件，保存到labelme文件中

# 这里是这些模型的下载路径
https://github.com/wkentaro/labelme/releases/download/sam-20230416/sam_vit_b_01ec64.quantized.encoder.onnx
https://github.com/wkentaro/labelme/releases/download/sam-20230416/sam_vit_b_01ec64.quantized.decoder.onnx
https://github.com/wkentaro/labelme/releases/download/sam-20230416/sam_vit_l_0b3195.quantized.encoder.onnx
https://github.com/wkentaro/labelme/releases/download/sam-20230416/sam_vit_l_0b3195.quantized.decoder.onnx
https://github.com/wkentaro/labelme/releases/download/sam-20230416/sam_vit_h_4b8939.quantized.encoder.onnx
https://github.com/wkentaro/labelme/releases/download/sam-20230416/sam_vit_h_4b8939.quantized.decoder.onnx
https://github.com/labelmeai/efficient-sam/releases/download/onnx-models-20231225/efficient_sam_vitt_encoder.onnx
https://github.com/labelmeai/efficient-sam/releases/download/onnx-models-20231225/efficient_sam_vitt_decoder.onnx
https://github.com/labelmeai/efficient-sam/releases/download/onnx-models-20231225/efficient_sam_vits_encoder.onnx
https://github.com/labelmeai/efficient-sam/releases/download/onnx-models-20231225/efficient_sam_vits_decoder.onnx

打开anaconda的envs目录下的此路径，打开-init-.py文件进行编辑

替换全部-init-.py文件的内容为以下代码，把MODEL_DIR改为第一步models文件的存放路径。

import os.path as osp
MODEL_DIR = "D:/Anaconda3/envs/data/Lib/site-packages/labelme/models/"
from .efficient_sam import EfficientSam
from .segment_anything_model import SegmentAnythingModel
from .text_to_annotation import get_rectangles_from_texts  # NOQA: F401
from .text_to_annotation import get_shapes_from_annotations  # NOQA: F401
from .text_to_annotation import non_maximum_suppression  # NOQA: F401


class SegmentAnythingModelVitB(SegmentAnythingModel):
    name = "SegmentAnything (speed)"

    def __init__(self):
        super().__init__(
            encoder_path=osp.join(MODEL_DIR, "sam_vit_b_01ec64.quantized.encoder.onnx"),  # NOQA
            decoder_path=osp.join(MODEL_DIR, "sam_vit_b_01ec64.quantized.decoder.onnx"),  # NOQA
        )


class SegmentAnythingModelVitL(SegmentAnythingModel):
    name = "SegmentAnything (balanced)"

    def __init__(self):
        super().__init__(
            encoder_path=osp.join(MODEL_DIR, "sam_vit_l_0b3195.quantized.encoder.onnx"),  # NOQA
            decoder_path=osp.join(MODEL_DIR, "sam_vit_l_0b3195.quantized.decoder.onnx"),  # NOQA
        )


class SegmentAnythingModelVitH(SegmentAnythingModel):
    name = "SegmentAnything (accuracy)"

    def __init__(self):
        super().__init__(
            encoder_path=osp.join(MODEL_DIR, "sam_vit_h_4b8939.quantized.encoder.onnx"),  # NOQA
            decoder_path=osp.join(MODEL_DIR, "sam_vit_h_4b8939.quantized.decoder.onnx"),  # NOQA
        )


class EfficientSamVitT(EfficientSam):
    name = "EfficientSam (speed)"

    def __init__(self):
        super().__init__(
            encoder_path=osp.join(MODEL_DIR, "efficient_sam_vitt_encoder.onnx"),  # NOQA
            decoder_path=osp.join(MODEL_DIR, "efficient_sam_vitt_decoder.onnx"),  # NOQA
        )


class EfficientSamVitS(EfficientSam):
    name = "EfficientSam (accuracy)"

    def __init__(self):
        super().__init__(
            encoder_path=osp.join(MODEL_DIR, "efficient_sam_vits_encoder.onnx"),  # NOQA
            decoder_path=osp.join(MODEL_DIR, "efficient_sam_vits_decoder.onnx"),  # NOQA
        )


MODELS = [
    SegmentAnythingModelVitB,
    SegmentAnythingModelVitL,
    SegmentAnythingModelVitH,
    EfficientSamVitT,
    EfficientSamVitS,
]