Skip to main content

Lvt evaluation of image object detection.

Project description

lvt-eval

  • 温馨提示:不要有中文路径!!!
  • lvt-eval代码库里包含【基础评估模块】

安装

  1. python>=3.8, (windows: need c++ env, https://airesources.oss-cn-hangzhou.aliyuncs.com/jkl/%E8%BE%B9%E7%BC%98%E5%8D%A1/VisualStudioSetup.exe)

  2. pip install -r requirements.txt -i https://pypi.tuna.tsinghua.edu.cn/simple


Lvt-eval基础评估模块 使用指南

调用studio aiport的评估方式

  1. 编写你的config文件存入 "configs/{你的config文件名}.json". ( e.g. config_cat.json )

例子:

(1) 多类别目标检测的配置文件:
{
    "label":[{"objectLabel":["烟"], "attrLabel":[], "id":0, "prediction":"smoke"},{"objectLabel":["火"], "attrLabel":[], "id":1, "prediction":"fire"}],
    "aiport": "http://192.1.2.238:8893/vql/v1/serving/process",
    "rawdata": "data/yh.json",
    "draw": false,
    "download": false,
    "raw_prediction_path": "prediction_dirs/raw_predictions_8893.json",
    "save_gt_coco": "gt_dirs/coco_groundtruth.json",
    "save_pred_path": "prediction_dirs/prediction_results.json"
}

(2) 多目标映射单目标检测的配置文件:
{
    "label":[{"objectLabel":["人"], "attrLabel":["躺", "趴"], "id":0, "prediction":"睡岗"}],
    "aiport": "http://192.1.2.238:8324/vql/v1/serving/process",
    "rawdata": "data/sleep_test_json_0621.json",
    "draw": false,
    "download": false,
    "raw_prediction_path": "prediction_dirs/raw_predictions_8324.json",
    "save_gt_coco": "gt_dirs/coco_groundtruth.json",
    "save_pred_path": "prediction_dirs/prediction_results.json"
}

(3) 单类别目标检测的配置文件:
{
    "label":[{"objectLabel":["person"], "attrLabel":[], "id":0, "prediction":"person"}],
    "aiport": "http://192.1.2.238:8312/vql/v1/serving/process",
    "rawdata": "data/xingren.json",
    "draw": false,
    "download": false,
    "raw_prediction_path": "prediction_dirs/raw_predictions_8312.json",
    "save_gt_coco": "gt_dirs/coco_groundtruth.json",
    "save_pred_path": "prediction_dirs/prediction_results.json"
}
==================================================================
参数描述:

"label":
    "objectLabel": studio-json中类别标签,用[str]表示,可添加多个类别 
    "attrLabel": studio-json中属性标签,用[str]表示,可添加多种属性
    "id": 默认从0开始,多个类别需按顺序添加 
    "prediction": 模型返回的类别标签
"aiport": 模型serving接口
"rawdata": studio-json格式的原数据
"draw": 默认为false, true表示绘制真实框和预测框
"download": 下载图片, 如果draw=True,则需要download=True
"raw_prediction_path": 保存接口的原始输出,文件名为“raw_predictions_{模型serving接口的端口号}.json”
"save_gt_coco": 将studio-json格式保存为coco-json真实标注格式
"save_pred_path": 将接口的原始输出保存为coco-json预测格式
  1. 在终端执行
python od_evaluator.py --mode studio_json --config {your config files} 

使用coco-json格式的评估

  1. 准备数据coco-json格式的groundtruth和prediction

  2. 在终端执行

python od_evaluator.py --mode coco_json --gt_json {groundtruth的路径}  --pred_json {prediction的路径} 

如何使用lvt-eval基础评估模块

  1. 使用 mode = studio aiport:
python od_evaluator.py --mode studio_json --config configs/config_hat.json
  1. 使用 mode = coco json:
python od_evaluator.py --mode coco_json --gt_json example_data/only_no_glove.json --pred_json example_data/only_no_glove_pred.json

输出返回值

[{
'threshold': 0.1,
'metrics': {
    'labels': [{
        'name': 'no_gloves',
        'AP': 0.44738234431641344,
        'AR': 0.9758064516129032,
        'P': 0.8539707196022313,
        'R': 0.9435483870967742,
        'AP50': 0.9208848985211565,
        'AR50': 0.5604838709677418
    }],
    'AP': 0.44738234431641344,
    'AP50': 0.9208848985211565,
    'AP75': 0.33124699279215397,
    'APs': 0.3775518845889323,
    'APm': 0.546416568798659,
    'APl': nan,
    'AR_50_95_all_1': 0.39919354838709686,
    'AR_50_95_all_10': 0.5604838709677418,
    'AR_50_95_all_100': 0.5604838709677418,
    'ARs': 0.5054054054054054,
    'ARm': 0.6419999999999999,
    'ARl': nan,
    'P': 0.8539707196022313,
    'R': 0.9435483870967742
}
}]

Project details


Download files

Download the file for your platform. If you're not sure which to choose, learn more about installing packages.

Source Distribution

Lvt-Evaluation-0.1.0.tar.gz (35.4 kB view details)

Uploaded Source

File details

Details for the file Lvt-Evaluation-0.1.0.tar.gz.

File metadata

  • Download URL: Lvt-Evaluation-0.1.0.tar.gz
  • Upload date:
  • Size: 35.4 kB
  • Tags: Source
  • Uploaded using Trusted Publishing? No
  • Uploaded via: twine/3.1.1 pkginfo/1.4.2 requests/2.22.0 setuptools/45.2.0 requests-toolbelt/0.8.0 tqdm/4.30.0 CPython/3.8.10

File hashes

Hashes for Lvt-Evaluation-0.1.0.tar.gz
Algorithm Hash digest
SHA256 1aa2c865ea13338df89c72fde6738654cc7598f76ccb149977fa7fdc90a022a7
MD5 cae0668496bb1924f74579ffb0ed7878
BLAKE2b-256 f8b957bde64f1db4bb7514063b462d1629ffc76d5f9ceb3933a2828696b69b2b

See more details on using hashes here.

Supported by

AWS AWS Cloud computing and Security Sponsor Datadog Datadog Monitoring Fastly Fastly CDN Google Google Download Analytics Microsoft Microsoft PSF Sponsor Pingdom Pingdom Monitoring Sentry Sentry Error logging StatusPage StatusPage Status page