标签：lb name RLE 标签 rle vec file np csv

# coding:utf-8
from glob import glob
import os
import SimpleITK as sitk
from pathlib import Path
import numpy as np
import imageio
import pandas as pd

def rle_encode(mask, bg = 0) -> dict:
    vec = mask.flatten()
    nb = len(vec)
    where = np.flatnonzero
    starts = np.r_[0, where(~np.isclose(vec[1:], vec[:-1], equal_nan=True)) + 2]
    lengths = np.diff(np.r_[starts, nb])
    values = vec[starts]
    assert len(starts) == len(lengths) == len(values)
    rle = {}
    for start, length, val in zip(starts, lengths, values):
        if val == bg:
            continue
        rle[val] = rle.get(val, []) + [str(start), length]
    # post-processing
    rle = {lb: " ".join(map(str, id_lens)) for lb, id_lens in rle.items()}
    return rle

def generate_rel(LABELS, path):
    preds = []

    for i in range(len(path)):
        file = path[i]
        file_name = file.split("\\")[-1].split("_seg")[0]
        case = file_name.split("_")[0]
        print("case:{}, file_name:{}".format(case, file_name))
        seg = sitk.ReadImage(file)
        seg = sitk.GetArrayFromImage(seg)
        for j in range(seg.shape[0]):
            if j>=0 and j<9:
                number = str(0)+str(0)+str(0)+str(j+1)
            elif j>=9 and j<99:
                number = str(0)+str(0) + str(j+1)
            else:
                number = str(0) + str(j+1)
            name = file_name+"_slice_"+number
            output = seg[j, ...]
            Snapshot_img = np.zeros(shape=(seg.shape[1],seg.shape[2],3), dtype=np.uint8)  # png设置为3通道
            Snapshot_img[:, :, 0][np.where(output == 1)] = 1   #我们也有3个标签，其中值分别为1,2,3，所以我们需要给每个标签都赋予不同的通道
            Snapshot_img[:, :, 1][np.where(output == 2)] = 1
            Snapshot_img[:, :, 2][np.where(output == 3)] = 1
            rle_lb = rle_encode(Snapshot_img[:, :, 0]) if np.sum(Snapshot_img[:, :, 0]) > 1 else {}
            rle_sb = rle_encode(Snapshot_img[:, :, 1]) if np.sum(Snapshot_img[:, :, 1]) > 1 else {}
            rle_sto = rle_encode(Snapshot_img[:, :, 2]) if np.sum(Snapshot_img[:, :, 2]) > 1 else {}
            index = (0,1,2)
            rel = [rle_lb, rle_sb, rle_sto]
            preds += [{"id": name, "class": lb, "predicted": rle.get(1, "")} for i, rle, lb in zip(index, rel, LABELS)]
        df_pred = pd.DataFrame(preds)
        df_pred.to_csv("submission.csv", index=False)

if __name__ == "__main__":
    pred_file = glob(r"D:\compation\kaggle\3D_preprocess\a\*")  # 获取到该文件夹下所有的标签（3D nii文件）
    LABELS = ("large_bowel", "small_bowel", "stomach")
    generate_rel(LABELS, pred_file)

结果：

标签：lb,name,RLE,标签,rle,vec,file,np,csv
来源： https://www.cnblogs.com/peixu/p/16230277.html

本站声明： 1. iCode9 技术分享网（下文简称本站）提供的所有内容，仅供技术学习、探讨和分享；
2. 关于本站的所有留言、评论、转载及引用，纯属内容发起人的个人观点，与本站观点和立场无关；
3. 关于本站的所有言论和文字，纯属内容发起人的个人观点，与本站观点和立场无关；
4. 本站文章均是网友提供，不完全保证技术分享内容的完整性、准确性、时效性、风险性和版权归属；如您发现该文章侵犯了您的权益，可联系我们第一时间进行删除；
5. 本站为非盈利性的个人网站，所有内容不会用来进行牟利，也不会利用任何形式的广告来间接获益，纯粹是为了广大技术爱好者提供技术内容和技术思想的分享性交流网站。