facebook 圖像比賽

Training a model to detect text from the ground up could be a very hard and frustrating task. Conventional way is to use R-CNN with Feature Pyramid Network or using algorithms like YOLO.

訓(xùn)練模型以從頭開始檢測文本可能是一項非常艱巨而令人沮喪的任務(wù)。 常規(guī)方法是將R-CNN與功能金字塔網(wǎng)絡(luò)配合使用，或使用類似YOLO的算法。

Either of them is very hard to implement if you are not aware of the mathematics and logic behind these.

如果您不了解它們背??后的數(shù)學(xué)和邏輯，則很難實現(xiàn)它們中的任何一個。

Detectron 2 which is developed by Facebook AI research team is a state-of-the-art object detection model which is based on mask-r-CNN benchmark. It is powered by non-other than Pytorch deep learning framework. Key feature includes

Facebook AI研究團(tuán)隊開發(fā)的Detectron 2是基于mask-r-CNN基準(zhǔn)的最新對象檢測模型。 它由Pytorch深度學(xué)習(xí)框架以外的其他組織提供支持。 主要功能包括

Source資源

Panoptic Segmentation: Another product of FAIR, is a type of segmentation That unifies the typically distinct tasks of semantic segmentation (assign a class label to each pixel) and instance segmentation (detect and segment each object instance).

全景分割 ：FAIR的另一種產(chǎn)品是一種分割類型，它將語義分割(將類標(biāo)簽分配給每個像素)和實例分割(檢測并分割每個對象實例)的典型任務(wù)統(tǒng)一起來。

FAIR GITHUB公平的Github

2. Dense pose: Used to map all human pixels of an RGB image to the 3D surface of the human body. This is powered by caffe2.

2. 密集姿勢 ：用于將RGB圖像的所有人類像素映射到人體的3D表面。 這是由caffe2驅(qū)動的。

“This model is meant to advance object detection by offering speedy training and addressing the issues companies face when making the step from research to production”

“此模型旨在通過提供快速培訓(xùn)并解決公司從研究到生產(chǎn)的過程中面臨的問題來提高對象檢測的效率”

讓我們開始吧！ (Let’s get started!)

Detectron 2 can be implemented for object detection using Google Colab Notebook. We are choosing Google Colab over local system to take advantage of GPU for faster training.

可以使用Google Colab Notebook將Detectron 2實施為對象檢測。 我們選擇Google Colab而非本地系統(tǒng)，以利用GPU進(jìn)行更快的培訓(xùn)。

步驟1：安裝和導(dǎo)入Detectron 2 (Step 1: Installing and Importing Detectron 2)

We will be writing this code on google colab or you get the entire notebook here.

我們將在google colab上編寫此代碼，或者您可以在此處獲取整個筆記本。

To get started we will be installing some dependencies such as COCO API,CUDA (To get info on GPU),Tourch Visison

首先，我們將安裝一些依賴項，例如COCO API，CUDA(有關(guān)GPU的信息)，Tourch Visison

Importing utilities and common libraries

導(dǎo)入實用程序和通用庫

步驟2：運行預(yù)訓(xùn)練的detectron2模型 (Step 2 : Run a pre-trained detectron2 model)

We will be using images from COCO dataset asnd will be running a pre trained model, If you want to run this model on a custome dataset see here.

我們將使用來自COCO數(shù)據(jù)集的圖像，然后將運行預(yù)先訓(xùn)練的模型。如果要在自定義數(shù)據(jù)集上運行此模型，請參見此處 。

Code for running model on image from COCO :

在COCO的圖像上運行模型的代碼：

…and you have successfuuly implemented your first project using detectron.

…您已經(jīng)成功地使用Detectron實施了您的第一個項目。

結(jié)果：可視化前后 (Results : Before and after visualization)

Source listed below.來源在下面列出。 Source資源

資源： (Resources :)

Facebook AI research page for detectron

Facebook針對AI的研究頁面

Paper explaining panoptic segmentation

解釋全景分割的論文

翻譯自: https://towardsdatascience.com/image-labelling-using-facebooks-detectron-4931e30c4d0c

facebook 圖像比賽

總結(jié)

以上是生活随笔為你收集整理的facebook 图像比赛_使用Facebook的Detectron进行图像标签的全部內(nèi)容，希望文章能夠幫你解決所遇到的問題。

如果覺得生活随笔網(wǎng)站內(nèi)容還不錯，歡迎將生活随笔推薦給好友。