数据集格式:labelme格式(不包含mask文件,仅仅包含jpg图片和对应的json文件)
图片数量(jpg文件个数):1164
标注数量(json文件个数):1164
标注类别数:37
标注类别名称:["number_plate","unauthorised_board","damaged_bus_stop","cleaning_required","damaged_rumble_strips","ravelling","unauthorised_opening","crack","faded_kerb","faded_marking","damaged_crash_barrier","encroachment","gantry_board","water_stagnation","face","damaged_road_stud","pothole","inappropriate_diversion","invisible_sign_board","faded_barrier_paint","damaged_blinker","plantation","excessive_vegetation","damaged_kerb","damaged_street_light","damaged_missing_manhole_covers","work_in_progress","damaged_footpath","rutting","hotspot","edge_drop","damaged_attenuator","damaged_anti_glare","damaged_delineator","damaged_sign_board","damaged_hazard_marker","rain_cuts"]
每个类别标注的框数:
number_plate(路名牌/标识牌)count = 132
unauthorised_board(违规广告牌)count = 29
damaged_bus_stop(公交站台损坏)count = 83
cleaning_required(需清洁)count = 17
damaged_rumble_strips(减速带损坏)count = 31
ravelling(松散剥落)count = 66
unauthorised_opening(私开道口)count = 44
crack(裂缝)count = 192
faded_kerb(路缘石褪色)count = 19
faded_marking(标线模糊)count = 37
damaged_crash_barrier(防撞护栏损坏)count = 60
encroachment(违章占道)count = 40
gantry_board(门架式标志牌)count = 22
water_stagnation(路面积水)count = 15
face(人脸)count = 9
damaged_road_stud(路钮/道钉损坏)count = 42
pothole(坑槽)count = 34
inappropriate_diversion(不当绕行)count = 18
invisible_sign_board(标志牌模糊不清)count = 38
faded_barrier_paint(护栏漆面褪色)count = 31
damaged_blinker(警示灯损坏)count = 22
plantation(绿化种植)count = 74
excessive_vegetation(植被过度生长)count = 2
damaged_kerb(路缘石损坏)count = 35
damaged_street_light(路灯损坏)count = 51
damaged_missing_manhole_covers(井盖损坏/缺失)count = 70
work_in_progress(施工中)count = 29
damaged_footpath(人行道损坏)count = 35
rutting(车辙)count = 20
hotspot(事故多发点)count = 9
edge_drop(路肩沉陷)count = 42
damaged_attenuator(缓冲垫损坏)count = 12
damaged_anti_glare(防眩板损坏)count = 150
damaged_delineator(轮廓标损坏)count = 41
damaged_sign_board(标志牌损坏)count = 34
damaged_hazard_marker(危险警示标志损坏)count = 9
rain_cuts(雨水冲刷沟)count = 7
总框数:1601
使用标注工具:labelme=5.5.0
所在github仓库:firc-dataset
图片分辨率:多分辨率图片,如1900x1069,1082x476等
标注规则:对类别进行画多边形框polygon
重要说明:可以将数据集用labelme打开编辑,json数据集需自己转成mask或者yolo格式或者coco格式作语义分割或者实例分割
特别声明:本数据集不对训练的模型或者权重文件精度作任何保证
图片预览:
标注例子:
原图(随机选16张图):
标注绘制结果:
labelme编辑图实例: