微软OmniParser项目

文摘   2024-10-29 12:45   韩国  

项目地址:https://github.com/microsoft/OmniParser

欢迎关注我的个人repo并点亮星星:

https://github.com/xinyuwei-david/david-share.git/Multimodal-Models/

OmniParser


Box Threshold 和 IOU Threshold 是用于控制目标检测模型(YOLO模型)输出的两个重要参数。它们的具体含义如下:

  1. Box Threshold(框阈值):

  • 这个参数用来设置检测到的边界框(bounding box)的置信度阈值。

  • 边界框置信度是模型对检测到的目标的置信程度。置信度越高,模型越确定这个边界框内包含一个目标。

  • Box Threshold 的值介于0.01到1.0之间。当检测到的边界框置信度低于这个阈值时,这个边界框将被过滤掉,不会出现在最终的输出结果中。

  • 例如,如果 Box Threshold 设置为0.05,那么置信度低于0.05的边界框将被忽略。

  • IOU Threshold(交并比阈值):

    • IOU(Intersection over Union)是用来衡量两个边界框之间重叠程度的指标。

    • IOU Threshold 用来设置在进行非极大值抑制(NMS, Non-Maximum Suppression)时的阈值。NMS 是一种后处理步骤,用于移除重叠的边界框,只保留最有可能的那个。

    • IOU Threshold 的值介于0.01到1.0之间。当两个边界框的IOU值大于这个阈值时,置信度较低的边界框将被移除。

    • 例如,如果 IOU Threshold 设置为0.1,那么当两个边界框的IOU值大于0.1时,置信度较低的那个边界框将被移除。

    总结

    • Box Threshold:用于过滤掉置信度低于该阈值的边界框。

    • IOU Threshold:用于在非极大值抑制过程中移除重叠度高于该阈值的边界框。

    这两个参数的设置可以帮助你调整目标检测的精度和召回率,以便在不同的应用场景中获得最佳的检测效果。

    Text Box ID 0: MenuText Box ID 1: OpenAI_s_o1_evaluatio_Text Box ID 2: CreateText Box ID 3: SignText Box ID 4: @iibText Box ID 5: All toolsText Box ID 6: EditText Box ID 7: ConvertText Box ID 8: Find text or toolsText Box ID 9: Al AssistantText Box ID 10: WinSCPText Box ID 11: PromptText Box ID 12: github-localText Box ID 13: SupposeText Box ID 14: mecicaText Box ID 15: following multiple-choiceText Box ID 16: quesions andText Box ID 17: anation?Text Box ID 18: 2Text Box ID 19: siued icevelaping ereasldairheText Box ID 20: Dou9ht She Rris prescribed Ocps Text Box ID 21: ereaotbackeyticcectoratText Box ID 22: eiectText Box ID 23: BaminationText Box ID 24: breast?Text Box ID 25: develText Box ID 26: stageText Box ID 27: and pubic hairText Box ID 28: minimal (TannerText Box ID 29: 31ag8Text Box ID 30: mostText Box ID 31: probable diagnosis?"Text Box ID 32: opaText Box ID 33: -"PrematneText Box ID 34: opb" :Text Box ID 35: syndrome"Text Box ID 36: MiillenanText Box ID 37: agenesisText Box ID 38: opd"Text Box ID 39: Androgen insensiivty"Text Box ID 40: oCamText Box ID 41: Q1-previewText Box ID 42: mostText Box ID 43: probable diagnosis forText Box ID 44: 6-year-oldText Box ID 45: Androgen Insensitivity SyndromeText Box ID 46: (OptionText Box ID 47: lanation:Text Box ID 48: most probable diagnosisText Box ID 49: et'3 analyzeText Box ID 50: keyText Box ID 51: features presented:Text Box ID 52: Amenorrhea:Text Box ID 53: hasText Box ID 54: started menstmuating by theText Box ID 55: Development:Text Box ID 56: StanedText Box ID 57: loping breastsText Box ID 58: and -Text Box ID 59: examination;Text Box ID 60: breasts areText Box ID 61: developed (TannerText Box ID 62: 4KSz#IhText Box ID 63: Minimal Pubic Hair: SheText Box ID 64: pubic hair (Tanner StageText Box ID 65: and Athleticism:Text Box ID 66: hasText Box ID 67: normal heightandText Box ID 68: football player; indicatingText Box ID 69: normaText Box ID 70: growth and development:Text Box ID 71: IneffectivenessText Box ID 72: ContraceptiveText Box ID 73: wa; prescnbedText Box ID 74: years 2g0Text Box ID 75: suggesting tat homonal therapy did not >Text Box ID 76: menstuation:Text Box ID 77: Interpreting the ClinicalText Box ID 78: DevelopmentText Box ID 79: This combinationText Box ID 80: that estrogen efectText Box ID 81: present (breast development), but androgen effectText Box ID 82: (lackText Box ID 83: pubic hair).Text Box ID 84: 8fa978fe7_Text Box ID 85: AmenorrheaText Box ID 86: NormalText Box ID 87: External FemalText Box ID 88: Genitalia andText Box ID 89: Uterus:Text Box ID 90: conditions,Text Box ID 91: mayText Box ID 92: aosent OrText Box ID 93: underdevelopedText Box ID 94: eadingText Box ID 95: primary amenorrhea despiteText Box ID 96: secondary sexual characenstics_Text Box ID 97: Differential Diagnoses:Text Box ID 98: evaluate each option basedText Box ID 99: te clinical features:Text Box ID 100: Premature OvarianText Box ID 101: Expected Findings:Text Box ID 102: CessationText Box ID 103: ovarian functionText Box ID 104: #hz# YAYText Box ID 105: Decreased estrogenText Box ID 106: t4txtText Box ID 107: LackText Box ID 108: breast de_Text Box ID 109: opmentText Box ID 110: regression_Text Box ID 111: 249Text Box ID 112: PresenceText Box ID 113: pubic andText Box ID 114: hairText Box ID 115: adrenal androgensText Box ID 116: unaffected)-Text Box ID 117: Conclusion: Unlikely; becauseText Box ID 118: paient has well-developed breasts andText Box ID 119: pubicText Box ID 120: 280Text Box ID 121: Syndrome (45,XOJ:Text Box ID 122: PDFText Box ID 123: OpenAIText Box ID 124: TIEIPText Box ID 125: 2024102_Text Box ID 126: 249Text Box ID 127: TE1B]Text Box ID 128: 2024102_Text Box ID 129: E-SignText Box ID 130: Stage 'Icon Box ID 131: Character Highlighting ColorIcon Box ID 132: Up arrowIcon Box ID 133: sending a message or email.Icon Box ID 134: HomeIcon Box ID 135: Add commentIcon Box ID 136: LinkIcon Box ID 137: SaveIcon Box ID 138: App launcherIcon Box ID 139: PrintIcon Box ID 140: RedoIcon Box ID 141: EmailIcon Box ID 142: CalendarIcon Box ID 143: NotificationsIcon Box ID 144: Help (Alt+H)Icon Box ID 145: a brick wall.Icon Box ID 146: minimizing a window.Icon Box ID 147: M0,0L9,0 4.5,5zIcon Box ID 148: CloseIcon Box ID 149: minimizing a window.Icon Box ID 150: CommentIcon Box ID 151: Copy

    本地目录结构:

    (omni) root@a100vm:~/OmniParser/weights# ls -al ./*-rw-r--r-- 1 root root  319 Oct 28 11:20 ./convert_safetensor_to_pt.py

    ./icon_caption_blip2:total 14628564drwxr-xr-x 2 root root 4096 Oct 28 13:40 .drwxr-xr-x 5 root root 4096 Oct 28 11:47 ..-rw-r--r-- 1 root root 1140 Oct 28 11:57 LICENSE-rw-r--r-- 1 root root 985 Oct 28 11:57 config.json-rw-r--r-- 1 root root 136 Oct 28 11:57 generation_config.json-rw-r--r-- 1 root root 4998300711 Oct 9 22:58 pytorch_model-00001-of-00002.bin-rw-r--r-- 1 root root 4998064248 Oct 24 14:31 pytorch_model-00001-of-00002.safetensors-rw-r--r-- 1 root root 2491516218 Oct 9 22:58 pytorch_model-00002-of-00002.bin-rw-r--r-- 1 root root 2491456448 Oct 24 14:31 pytorch_model-00002-of-00002.safetensors-rw-r--r-- 1 root root 121726 Oct 28 12:01 pytorch_model.bin.index.json


    ./icon_caption_florence:total 1058612drwxr-xr-x 2 root root 4096 Oct 28 11:56 .drwxr-xr-x 5 root root 4096 Oct 28 11:47 ..-rw-r--r-- 1 root root 72204 Oct 28 11:40 LICENSE-rw-r--r-- 1 root root 5663 Oct 28 11:54 config.json-rw-r--r-- 1 root root 292 Oct 28 11:54 generation_config.json-rw-r--r-- 1 root root 1083916964 Oct 25 23:19 model.safetensors

    ./icon_detect:total 18276drwxr-xr-x 2 root root 4096 Oct 28 11:52 .drwxr-xr-x 5 root root 4096 Oct 28 11:47 ..-rw-r--r-- 1 root root 400264 Oct 28 11:43 LICENSE-rw-r--r-- 1 root root 12222450 Oct 28 13:45 best.pt-rw-r--r-- 1 root root 6075790 Oct 24 23:11 model.safetensors-rw-r--r-- 1 root root 1087 Oct 28 11:52 model.yaml

    在A100上推理的的话,肯定是很快的:

    通过笔记本的CPU也能运行,但比GPU的运行速度要慢很多。

    大魏分享
    https://github.com/davidsajare/david-share.git
     最新文章